Publications by our Members

Remote patient monitoring (RPM) is the use of digital technologies to improve patient care at a distance. However, current RPM solutions are often biased toward tech-savvy patients. To foster health equity, researchers have studied how to address the socio-economic and cognitive needs of diverse patient groups, but their emotional needs have remained largely neglected. We perform the first qualitative study to explore the emotional needs of diverse patients around RPM. Specifically, we conduct a thematic analysis of 18 interviews and 4 focus groups at a large US healthcare organization. We identify emotional needs that lead to four emotional tensions within and across stakeholder groups when applying an equity focus to the design and implementation of RPM technologies. The four emotional tensions are making diverse patients feel: (i) heard vs. exploited; (ii) seen vs. deprioritized for efficiency; (iii) empowered vs. anxious; and (iv) cared for vs. detached from care. To manage these emotional tensions across stakeholders, we develop design recommendations informed by a paradox mindset (i.e., ‘both-and’ rather than ‘and-or’ strategies).

Recent works on the global place recognition treat the task as a retrieval problem, where an off-the-shelf global descriptor is commonly designed in image-based and LiDAR-based modalities. However, it is non-trivial to perform accurate image-LiDAR global place recognition since extracting consistent and robust global descriptors from different domains (2D images and 3D point clouds) is challenging. To address this issue, we propose a novel Voxel-Cross-Pixel (VXP) approach, which establishes voxel and pixel correspondences in a self-supervised manner and brings them into a shared feature space. Specifically, VXP is trained in a two-stage manner that first explicitly exploits local feature correspondences and enforces similarity of global descriptors. Extensive experiments on the three benchmarks (Oxford RobotCar, ViViD++ and KITTI) demonstrate our method surpasses the state-of-the-art cross-modal retrieval by a large margin.

Crime is responsible for major financial losses and serious harm to the well-being of individuals, and, hence, a crucial task of police operations is effective patrolling. Yet, in existing decision models aimed at police operations, microscopic routing decisions from patrolling are not considered, and, furthermore, the objective is limited to surrogate metrics (e. g., response time) instead of crime prevention. In this paper, we thus formalize the decision problem of dynamic police patrolling as a Markov decision process that models microscopic routing decisions, so that the expected number of prevented crimes are maximized. We experimentally show that standard solution approaches for our decision problem are not scalable to real-world settings. As a remedy, we present a tailored and highly efficient Monte Carlo tree search algorithm. We then demonstrate our algorithm numerically using real-world crime data from Chicago and show that the decision-making by our algorithm offers significant improvements for crime prevention over patrolling tactics from current practice. Informed by our results, we finally discuss implications for improving the patrolling tactics in police operations.

Decoding strategies for large language models (LLMs) are a critical but often underexplored aspect of text generation tasks. Since LLMs produce probability distributions over the entire vocabulary, various decoding methods have been developed to transform these probabilities into coherent and fluent text, each with its own set of hyperparameters. In this study, we present a large-scale, comprehensive analysis of how hyperparameter selection affects text quality in open-ended text generation across multiple LLMs, datasets, and evaluation metrics. Through an extensive sensitivity analysis, we provide practical guidelines for hyperparameter tuning and demonstrate the substantial influence of these choices on text quality. Using three established datasets, spanning factual domains (e.g., news) and creative domains (e.g., fiction), we show that hyperparameter tuning significantly impacts generation quality, though its effects vary across models and tasks. We offer in-depth insights into these effects, supported by both human evaluations and a synthesis of widely-used automatic evaluation metrics.

A large amount of local and culture-specific knowledge (e.g., people, traditions, food) can only be found in documents written in dialects. While there has been extensive research conducted on cross-lingual information retrieval (CLIR), the field of cross-dialect retrieval (CDIR) has received limited attention. Dialect retrieval poses unique challenges due to the limited availability of resources to train retrieval models and the high variability in non-standardized languages. We study these challenges on the example of German dialects and introduce the first German dialect retrieval dataset, dubbed WikiDIR, which consists of seven German dialects extracted from Wikipedia. Using WikiDIR, we demonstrate the weakness of lexical methods in dealing with high lexical variation in dialects. We further show that commonly used zero-shot cross-lingual transfer approach with multilingual encoders do not transfer well to extremely low-resource setups, motivating the need for resource-lean and dialect-specific retrieval models. We finally demonstrate that (document) translation is an effective way to reduce the dialect gap in CDIR.

Transliterating related languages that use different scripts into a common script shows effectiveness in improving crosslingual transfer in downstream tasks. However, this methodology often makes pretraining a model from scratch unavoidable, as transliteration brings about new subwords not covered in existing multilingual pretrained language models (mPLMs). This is not desired because it takes a lot of computation budget for pretraining. A more promising way is to make full use of available mPLMs. To this end, this paper proposes a simple but effective framework: Transliterate-Merge-Initialize (TransMI), which can create a strong baseline well-suited for data that is transliterated into a common script by exploiting an mPLM and its accompanied tokenizer. TransMI has three stages: (a) transliterate the vocabulary of an mPLM into a common script; (b) merge the new vocabulary with the original vocabulary; and (c) initialize the embeddings of the new subwords. We applied TransMI to three recent strong mPLMs, and our experiments demonstrate that TransMI not only preserves their ability to handle non-transliterated data, but also enables the models to effectively process transliterated data: the results show a consistent improvement of 3% to 34%, varying across different models and tasks.

Recent studies have shown that post-aligning multilingual pretrained language models (mPLMs) using alignment objectives on both original and transliterated data can improve crosslingual alignment. This improvement further leads to better crosslingual transfer performance. However, it remains unclear how and why a better crosslingual alignment is achieved, as this technique only involves transliterations, and does not use any parallel data. This paper attempts to explicitly evaluate the crosslingual alignment and identify the key elements in transliteration-based approaches that contribute to better performance. For this, we train multiple models under varying setups for two pairs of related languages: (1) Polish and Ukrainian and (2) Hindi and Urdu. To assess alignment, we define four types of similarities based on sentence representations. Our experiments show that adding transliterations alone improves the overall similarities, even for random sentence pairs. With the help of auxiliary alignment objectives, especially the contrastive objective, the model learns to distinguish matched from random pairs, leading to better alignments. However, we also show that better alignment does not always yield better downstream performance, suggesting that further research is needed to clarify the connection between alignment and performance.

We explore the potential of pixel-based models for transfer learning from standard languages to dialects. These models convert text into images that are divided into patches, enabling a continuous vocabulary representation that proves especially useful for out-of-vocabulary words common in dialectal data. Using German as a case study, we compare the performance of pixel-based models to token-based models across various syntactic and semantic tasks. Our results show that pixel-based models outperform token-based models in part-of-speech tagging, dependency parsing and intent detection for zero-shot dialect evaluation by up to 26 percentage points in some scenarios, though not in Standard German. However, pixel-based models fall short in topic classification. These findings emphasize the potential of pixel-based models for handling dialectal data, though further research should be conducted to assess their effectiveness in various linguistic contexts.

Orthographic variation is very common in Luxembourgish texts due to the absence of a fully-fledged standard variety. Additionally, developing NLP tools for Luxembourgish is a difficult task given the lack of annotated and parallel data, which is exacerbated by ongoing standardization. In this paper, we propose the first sequence-to-sequence normalization models using the ByT5 and mT5 architectures with training data obtained from word-level real-life variation data. We perform a fine-grained, linguistically-motivated evaluation to test byte-based, word-based and pipeline-based models for their strengths and weaknesses in text normalization. We show that our sequence model using real-life variation data is an effective approach for tailor-made normalization in Luxembourgish.

Many systems occurring in real-world applications, such as controlling the motions of robots or modeling the spread of diseases, are switched impulsive systems. To ensure that the system state stays in a safe region (e.g., to avoid collisions with obstacles), barrier functions are widely utilized. As the system dimension increases, deriving suitable barrier functions becomes extremely complex. Fortunately, many systems consist of multiple subsystems, such as different areas where the disease occurs. In this work, we present sufficient conditions for interconnected switched impulsive systems to maintain safety by constructing local barrier functions for the individual subsystems instead of a global one, allowing for much easier and more efficient derivation. To validate our results, we numerically demonstrate its effectiveness using an epidemiological model.

We study the problem of monitoring machine learning models under temporal distribution shifts, where circumstances change gradually over time, often leading to unnoticed yet significant declines in accuracy. We propose Incremental Uncertainty-aware Performance Monitoring (IUPM), a novel label-free method that estimates model performance by modeling time-dependent shifts using optimal transport. IUPM also quantifies uncertainty in performance estimates and introduces an active labeling strategy to reduce this uncertainty. We further showcase the benefits of IUPM on different datasets and simulated temporal shifts over existing baselines.

We show that variational learning can significantly improve the accuracy and calibration of Low-Rank Adaptation (LoRA) without a substantial increase in the cost. We replace AdamW by the Improved Variational Online Newton (IVON) algorithm to finetune large language models. For Llama-2 with 7 billion parameters, IVON improves the accuracy over AdamW by 2.8% and expected calibration error by 4.6%. The accuracy is also better than the other Bayesian alternatives, yet the cost is lower and the implementation is easier. Our work provides additional evidence for the effectiveness of IVON for large language models.

Hierarchical clustering has usually been addressed by discrete optimization using heuristics or continuous optimization of relaxed scores for hierarchies. In this work, we propose to optimize expected scores under a probabilistic model over hierarchies. (1) We show theoretically that the global optimal values of the expected Dasgupta cost and Tree-Sampling divergence (TSD), two unsupervised metrics for hierarchical clustering, are equal to the optimal values of their discrete counterparts contrary to some relaxed scores. (2) We propose Expected Probabilistic Hierarchies (EPH), a probabilistic model to learn hierarchies in data by optimizing expected scores. EPH uses differentiable hierarchy sampling enabling end-to-end gradient descent based optimization, and an unbiased subgraph sampling approach to scale to large datasets. (3) We evaluate EPH on synthetic and real-world datasets including vector and graph datasets. EPH outperforms all other approaches quantitatively and provides meaningful hierarchies in qualitative evaluations.

Estimating causal quantities from observational data is crucial for understanding the safety and effectiveness of medical treatments. However, to make reliable inferences, medical practitioners require not only estimating averaged causal quantities, such as the conditional average treatment effect, but also understanding the randomness of the treatment effect as a random variable. This randomness is referred to as aleatoric uncertainty and is necessary for understanding the probability of benefit from treatment or quantiles of the treatment effect. Yet, the aleatoric uncertainty of the treatment effect has received surprisingly little attention in the causal machine learning community. To fill this gap, we aim to quantify the aleatoric uncertainty of the treatment effect at the individualized (covariate-conditional) level, namely, the conditional distribution of the treatment effect (CDTE). Unlike average causal quantities, the CDTE is not point identifiable without strong additional assumptions. As a remedy, we employ partial identification to obtain sharp bounds on the CDTE and thereby quantify the aleatoric uncertainty of the treatment effect. We then develop a novel, orthogonal learner for the bounds on the CDTE, which we call AU-learner. We further show that our AU-learner has several strengths in that it satisfies Neyman-orthogonality and is doubly robust. Finally, we propose a fully-parametric deep learning instantiation of our AU-learner.

Scientific hypotheses typically concern specific aspects of complex, imperfectly understood or entirely unknown mechanisms, such as the effect of gene expression levels on phenotypes or how microbial communities influence environmental health. Such queries are inherently causal (rather than purely associational), but in many settings, experiments can not be conducted directly on the target variables of interest, but are indirect. Therefore, they perturb the target variable, but do not remove potential confounding factors. If, additionally, the resulting experimental measurements are multi-dimensional and the studied mechanisms nonlinear, the query of interest is generally not identified. We develop an adaptive strategy to design indirect experiments that optimally inform a targeted query about the ground truth mechanism in terms of sequentially narrowing the gap between an upper and lower bound on the query. While the general formulation consists of a bi-level optimization procedure, we derive an efficiently estimable analytical kernel-based estimator of the bounds for the causal effect, a query of key interest, and demonstrate the efficacy of our approach in confounded, multivariate, nonlinear synthetic settings.

Neural network sparsification is a promising avenue to save computational time and memory costs, especially in an age where many successful AI models are becoming too large to naïvely deploy on consumer hardware. While much work has focused on different weight pruning criteria, the overall sparsifiability of the network, i.e., its capacity to be pruned without quality loss, has often been overlooked. We present Sparsifiability via the Marginal likelihood (SpaM), a pruning framework that highlights the effectiveness of using the Bayesian marginal likelihood in conjunction with sparsity-inducing priors for making neural networks more sparsifiable. Our approach implements an automatic Occam’s razor that selects the most sparsifiable model that still explains the data well, both for structured and unstructured sparsification. In addition, we demonstrate that the pre-computed posterior Hessian approximation used in the Laplace approximation can be re-used to define a cheap pruning criterion, which outperforms many existing (more expensive) approaches. We demonstrate the effectiveness of our framework, especially at high sparsity levels, across a range of different neural network architectures and datasets.

Text-to-Image (T2I) models have made significant advancements in recent years, but they still struggle to accurately capture intricate details specified in complex compositional prompts. While fine-tuning T2I models with reward objectives has shown promise, it suffers from ‘reward hacking’ and may not generalize well to unseen prompt distributions. In this work, we propose Reward-based Noise Optimization (ReNO), a novel approach that enhances T2I models at inference by optimizing the initial noise based on the signal from one or multiple human preference reward models. Remarkably, solving this optimization problem with gradient ascent for 50 iterations yields impressive results on four different one-step models across two competitive benchmarks, T2I-CompBench and GenEval. Within a computational budget of 20-50 seconds, ReNO-enhanced one-step models consistently surpass the performance of all current open-source Text-to-Image models. Extensive user studies demonstrate that our model is preferred nearly twice as often compared to the popular SDXL model and is on par with the proprietary Stable Diffusion 3 with 8B parameters. Moreover, given the same computational resources, a ReNO-optimized one-step model outperforms widely-used open-source models such as SDXL and PixArt-α, highlighting the efficiency and effectiveness of ReNO in enhancing T2I model performance at inference time.

Uncertainty quantification (UQ) is a crucial but challenging task in many high-dimensional regression or learning problems to increase the confidence of a given predictor. We develop a new data-driven approach for UQ in regression that applies both to classical regression approaches such as the LASSO as well as to neural networks. One of the most notable UQ techniques is the debiased LASSO, which modifies the LASSO to allow for the construction of asymptotic confidence intervals by decomposing the estimation error into a Gaussian and an asymptotically vanishing bias component. However, in real-world problems with finite-dimensional data, the bias term is often too significant to be neglected, resulting in overly narrow confidence intervals. Our work rigorously addresses this issue and derives a data-driven adjustment that corrects the confidence intervals for a large class of predictors by estimating the means and variances of the bias terms from training data, exploiting high-dimensional concentration phenomena. This gives rise to non-asymptotic confidence intervals, which can help avoid overestimating uncertainty in critical applications such as MRI diagnosis. Importantly, our analysis extends beyond sparse regression to data-driven predictors like neural networks, enhancing the reliability of model-based deep learning. Our findings bridge the gap between established theory and the practical applicability of such debiased methods.

Credal sets are sets of probability distributions that are considered as candidates for an imprecisely known ground-truth distribution. In machine learning, they have recently attracted attention as an appealing formalism for uncertainty representation, in particular due to their ability to represent both the aleatoric and epistemic uncertainty in a prediction. However, the design of methods for learning credal set predictors remains a challenging problem. In this paper, we make use of conformal prediction for this purpose. More specifically, we propose a method for predicting credal sets in the classification task, given training data labeled by probability distributions. Since our method inherits the coverage guarantees of conformal prediction, our conformal credal sets are guaranteed to be valid with high probability (without any assumptions on model or distribution). We demonstrate the applicability of our method to natural language inference, a highly ambiguous natural language task where it is common to obtain multiple annotations per example.

The need for large text corpora has increased with the advent of pretrained language models and, in particular, the discovery of scaling laws for these models. Most available corpora have sufficient data only for languages with large dominant communities. However, there is no corpus available that (i) covers a wide range of minority languages; (ii) is generated by an open-source reproducible pipeline; and (iii) is rigorously cleaned from noise, making it trustworthy to use. We present GlotCC, a clean, document-level, 2TB general domain corpus derived from CommonCrawl, covering more than 1000 languages. We make GlotCC and the system used to generate it - including the pipeline, language identification model, and filters - available to the research community.

We introduce the Autoregressive PDE Emulator Benchmark (APEBench), a comprehensive benchmark suite to evaluate autoregressive neural emulators for solving partial differential equations. APEBench is based on JAX and provides a seamlessly integrated differentiable simulation framework employing efficient pseudo-spectral methods, enabling 46 distinct PDEs across 1D, 2D, and 3D. Facilitating systematic analysis and comparison of learned emulators, we propose a novel taxonomy for unrolled training and introduce a unique identifier for PDE dynamics that directly relates to the stability criteria of classical numerical methods. APEBench enables the evaluation of diverse neural architectures, and unlike existing benchmarks, its tight integration of the solver enables support for differentiable physics training and neural-hybrid emulators. Moreover, APEBench emphasizes rollout metrics to understand temporal generalization, providing insights into the long-term behavior of emulating PDE dynamics. In several experiments, we highlight the similarities between neural emulators and numerical simulators.

In many applications, we desire neural networks to exhibit invariance or equivariance to certain groups due to symmetries inherent in the data. Recently, frame-averaging methods emerged to be a unified framework for attaining symmetries efficiently by averaging over input-dependent subsets of the group, i.e., frames. What we currently lack is a principled understanding of the design of frames. In this work, we introduce a canonicalization perspective that provides an essential and complete view of the design of frames. Canonicalization is a classic approach for attaining invariance by mapping inputs to their canonical forms. We show that there exists an inherent connection between frames and canonical forms. Leveraging this connection, we can efficiently compare the complexity of frames as well as determine the optimality of certain frames. Guided by this principle, we design novel frames for eigenvectors that are strictly superior to existing methods – some are even optimal – both theoretically and empirically. The reduction to the canonicalization perspective further uncovers equivalences between previous methods. These observations suggest that canonicalization provides a fundamental understanding of existing frame-averaging methods and unifies existing equivariant and invariant learning methods.

Predicting potential outcomes of interventions from observational data is crucial for decision-making in medicine, but the task is challenging due to the fundamental problem of causal inference. Existing methods are largely limited to point estimates of potential outcomes with no uncertain quantification; thus, the full information about the distributions of potential outcomes is typically ignored. In this paper, we propose a novel causal diffusion model called DiffPO, which is carefully designed for reliable inferences in medicine by learning the distribution of potential outcomes. In our DiffPO, we leverage a tailored conditional denoising diffusion model to learn complex distributions, where we address the selection bias through a novel orthogonal diffusion loss. Another strength of our DiffPO method is that it is highly flexible (e.g., it can also be used to estimate different causal quantities such as CATE). Across a wide range of experiments, we show that our method achieves state-of-the-art performance.

Originally rooted in game theory, the Shapley Value (SV) has recently become an important tool in machine learning research. Perhaps most notably, it is used for feature attribution and data valuation in explainable artificial intelligence. Shapley Interactions (SIs) naturally extend the SV and address its limitations by assigning joint contributions to groups of entities, which enhance understanding of black box machine learning models. Due to the exponential complexity of computing SVs and SIs, various methods have been proposed that exploit structural assumptions or yield probabilistic estimates given limited resources. In this work, we introduce shapiq, an open-source Python package that unifies state-of-the-art algorithms to efficiently compute SVs and any-order SIs in an application-agnostic framework. Moreover, it includes a benchmarking suite containing 11 machine learning applications of SIs with pre-computed games and ground-truth values to systematically assess computational performance across domains. For practitioners, shapiq is able to explain and visualize any-order feature interactions in predictions of models, including vision transformers, language models, as well as XGBoost and LightGBM with TreeSHAP-IQ. With shapiq, we extend shap beyond feature attributions and consolidate the application of SVs and SIs in machine learning that facilitates future research.

Hyperparameter optimization is crucial for obtaining peak performance of machine learning models. The standard protocol evaluates various hyperparameter configurations using a resampling estimate of the generalization error to guide optimization and select a final hyperparameter configuration. Without much evidence, paired resampling splits, i.e., either a fixed train-validation split or a fixed cross-validation scheme, are often recommended. We show that, surprisingly, reshuffling the splits for every configuration often improves the final model’s generalization performance on unseen data. Our theoretical analysis explains how reshuffling affects the asymptotic behavior of the validation loss surface and provides a bound on the expected regret in the limiting regime. This bound connects the potential benefits of reshuffling to the signal and noise characteristics of the underlying optimization problem. We confirm our theoretical results in a controlled simulation study and demonstrate the practical usefulness of reshuffling in a large-scale, realistic hyperparameter optimization experiment. While reshuffling leads to test performances that are competitive with using fixed splits, it drastically improves results for a single train-validation holdout protocol and can often make holdout become competitive with standard CV while being computationally cheaper.

We introduce r-loopy Weisfeiler-Leman (r-ℓWL), a novel hierarchy of graph isomorphism tests and a corresponding GNN framework, r-ℓMPNN, that can count cycles up to length r+2. Most notably, we show that r-ℓWL can count homomorphisms of cactus graphs. This strictly extends classical 1-WL, which can only count homomorphisms of trees and, in fact, is incomparable to k-WL for any fixed k. We empirically validate the expressive and counting power of the proposed r-ℓMPNN on several synthetic datasets and present state-of-the-art predictive performance on various real-world datasets.

Multimodal foundation models serve numerous applications at the intersection of vision and language. Still, despite being pretrained on extensive data, they become outdated over time. To keep models updated, research into continual pretraining mainly explores scenarios with either (1) infrequent, indiscriminate updates on large-scale new data, or (2) frequent, sample-level updates. However, practical model deployment often operates in the gap between these two limit cases, as real-world applications often demand adaptation to specific subdomains, tasks or concepts – spread over the entire, varying life cycle of a model. In this work, we complement current perspectives on continual pretraining through a research test bed as well as provide comprehensive guidance for effective continual model updates in such scenarios. We first introduce FoMo-in-Flux, a continual multimodal pretraining benchmark with realistic compute constraints and practical deployment requirements, constructed over 63 datasets with diverse visual and semantic coverage. Using FoMo-in-Flux, we explore the complex landscape of practical continual pretraining through multiple perspectives: (1) A data-centric investigation of data mixtures and stream orderings that emulate real-world deployment situations, (2) a method-centric investigation ranging from simple fine-tuning and traditional continual learning strategies to parameter-efficient updates and model merging, (3) meta learning rate schedules and mechanistic design choices, and (4) the influence of model and compute scaling. Together, our insights provide a practitioner’s guide to continual multimodal pretraining for real-world deployment.

Semi-structured networks (SSNs) merge the structures familiar from additive models with deep neural networks, allowing the modeling of interpretable partial feature effects while capturing higher-order non-linearities at the same time. A significant challenge in this integration is maintaining the interpretability of the additive model component. Inspired by large-scale biomechanics datasets, this paper explores extending SSNs to functional data. Existing methods in functional data analysis are promising but often not expressive enough to account for all interactions and non-linearities and do not scale well to large datasets. Although the SSN approach presents a compelling potential solution, its adaptation to functional data remains complex. In this work, we propose a functional SSN method that retains the advantageous properties of classical functional regression approaches while also improving scalability. Our numerical experiments demonstrate that this approach accurately recovers underlying signals, enhances predictive performance, and performs favorably compared to competing methods.

Continuous action spaces in reinforcement learning (RL) are commonly defined as multidimensional intervals. While intervals usually reflect the action boundaries for tasks well, they can be challenging for learning because the typically large global action space leads to frequent exploration of irrelevant actions. Yet, little task knowledge can be sufficient to identify significantly smaller state-specific sets of relevant actions. Focusing learning on these relevant actions can significantly improve training efficiency and effectiveness. In this paper, we propose to focus learning on the set of relevant actions and introduce three continuous action masking methods for exactly mapping the action space to the state-dependent set of relevant actions. Thus, our methods ensure that only relevant actions are executed, enhancing the predictability of the RL agent and enabling its use in safety-critical applications. We further derive the implications of the proposed methods on the policy gradient. Using proximal policy optimization (PPO), we evaluate our methods on four control tasks, where the relevant action set is computed based on the system dynamics and a relevant state set. Our experiments show that the three action masking methods achieve higher final rewards and converge faster than the baseline without action masking.

Controllable text-to-image (T2I) diffusion models have shown impressive performance in generating high-quality visual content through the incorporation of various conditions. Current methods, however, exhibit limited performance when guided by skeleton human poses, especially in complex pose conditions such as side or rear perspectives of human figures. To address this issue, we present Stable-Pose, a novel adapter model that introduces a coarse-to-fine attention masking strategy into a vision Transformer (ViT) to gain accurate pose guidance for T2I models. Stable-Pose is designed to adeptly handle pose conditions within pre-trained Stable Diffusion, providing a refined and efficient way of aligning pose representation during image synthesis. We leverage the query-key self-attention mechanism of ViTs to explore the interconnections among different anatomical parts in human pose skeletons. Masked pose images are used to smoothly refine the attention maps based on target pose-related features in a hierarchical manner, transitioning from coarse to fine levels. Additionally, our loss function is formulated to allocate increased emphasis to the pose region, thereby augmenting the model’s precision in capturing intricate pose details. We assessed the performance of Stable-Pose across five public datasets under a wide range of indoor and outdoor human pose scenarios. Stable-Pose achieved an AP score of 57.1 in the LAION-Human dataset, marking around 13% improvement over the established technique ControlNet.

Contrastive learning has been a leading paradigm for self-supervised learning, but it is widely observed that it comes at the price of sacrificing useful features (eg colors) by being invariant to data augmentations. Given this limitation, there has been a surge of interest in equivariant self-supervised learning (E-SSL) that learns features to be augmentation-aware. However, even for the simplest rotation prediction method, there is a lack of rigorous understanding of why, when, and how E-SSL learns useful features for downstream tasks. To bridge this gap between practice and theory, we establish an information-theoretic perspective to understand the generalization ability of E-SSL. In particular, we identify a critical explaining-away effect in E-SSL that creates a synergy between the equivariant and classification tasks. This synergy effect encourages models to extract class-relevant features to improve its equivariant prediction, which, in turn, benefits downstream tasks requiring semantic features. Based on this perspective, we theoretically analyze the influence of data transformations and reveal several principles for practical designs of E-SSL. Our theory not only aligns well with existing E-SSL methods but also sheds light on new directions by exploring the benefits of model equivariance. We believe that a theoretically grounded understanding on the role of equivariance would inspire more principled and advanced designs in this field.

Going beyond mimicking limited human experiences, recent studies show initial evidence that, like humans, large language models (LLMs) are capable of improving their abilities purely by self-correction, i.e., correcting previous responses through self-examination, in certain circumstances. Nevertheless, little is known about how such capabilities arise. In this work, based on a simplified setup akin to an alignment task, we theoretically analyze self-correction from an in-context learning perspective, showing that when LLMs give relatively accurate self-examinations as rewards, they are capable of refining responses in an in-context way. Notably, going beyond previous theories on over-simplified linear transformers, our theoretical construction underpins the roles of several key designs of realistic transformers for self-correction: softmax attention, multi-head attention, and the MLP block. We validate these findings extensively on synthetic datasets. Inspired by these findings, we also illustrate novel applications of self-correction, such as defending against LLM jailbreaks, where a simple self-correction step does make a large difference. We believe that these findings will inspire further research on understanding, exploiting, and enhancing self-correction for building better foundation models.

Allocation tasks represent a class of problems where a limited amount of resources must be allocated to a set of entities at each time step. Prominent examples of this task include portfolio optimization or distributing computational workloads across servers. Allocation tasks are typically bound by linear constraints describing practical requirements that have to be strictly fulfilled at all times. In portfolio optimization, for example, investors may be obligated to allocate less than 30% of the funds into a certain industrial sector in any investment period. Such constraints restrict the action space of allowed allocations in intricate ways, which makes learning a policy that avoids constraint violations difficult. In this paper, we propose a new method for constrained allocation tasks based on an autoregressive process to sequentially sample allocations for each entity. In addition, we introduce a novel de-biasing mechanism to counter the initial bias caused by sequential sampling. We demonstrate the superior performance of our approach compared to a variety of Constrained Reinforcement Learning (CRL) methods on three distinct constrained allocation tasks: portfolio optimization, computational workload distribution, and a synthetic allocation benchmark.

In this work we design graph neural network architectures that capture optimal approximation algorithms for a large class of combinatorial optimization problems, using powerful algorithmic tools from semidefinite programming (SDP). Concretely, we prove that polynomial-sized message-passing algorithms can represent the most powerful polynomial time algorithms for Max Constraint Satisfaction Problems assuming the Unique Games Conjecture. We leverage this result to construct efficient graph neural network architectures, OptGNN, that obtain high-quality approximate solutions on landmark combinatorial optimization problems such as Max-Cut, Min-Vertex-Cover, and Max-3-SAT. Our approach achieves strong empirical results across a wide range of real-world and synthetic datasets against solvers and neural baselines. Finally, we take advantage of OptGNN’s ability to capture convex relaxations to design an algorithm for producing bounds on the optimal solution from the learned embeddings of OptGNN.

Overparametrized transformer networks are the state-of-the-art architecture for Large Language Models (LLMs). However, such models contain billions of parameters making large compute a necessity, while raising environmental concerns. To address these issues, we propose FinerCut, a new form of fine-grained layer pruning, which in contrast to prior work at the transformer block level, considers all self-attention and feed-forward network (FFN) layers within blocks as individual pruning candidates. FinerCut prunes layers whose removal causes minimal alternation to the model’s output – contributing to a new, lean, interpretable, and task-agnostic pruning method. Tested across 9 benchmarks, our approach retains 90% performance of Llama3-8B with 25% layers removed, and 95% performance of Llama3-70B with 30% layers removed, all without fine-tuning or post-pruning reconstruction. Strikingly, we observe intriguing results with FinerCut: 42% (34 out of 80) of the self-attention layers in Llama3-70B can be removed while preserving 99% of its performance – without additional fine-tuning after removal. Moreover, FinerCut provides a tool to inspect the types and locations of pruned layers, allowing to observe interesting pruning behaviors. For instance, we observe a preference for pruning self-attention layers, often at deeper consecutive decoder layers. We hope our insights inspire future efficient LLM architecture designs.

Prior-data fitted networks (PFNs), especially TabPFN, have shown significant promise in tabular data prediction. However, their scalability is limited by the quadratic complexity of the transformer architecture’s attention across training points. In this work, we propose a method to localize TabPFN, which embeds data points into a learned representation and performs nearest neighbor selection in this space. We evaluate it across six datasets, demonstrating its superior performance over standard TabPFN when scaling to larger datasets. We also explore its design choices and analyze the bias-variance trade-off of this localization method, showing that it reduces bias while maintaining manageable variance. This work opens up a pathway for scaling TabPFN to arbitrarily large tabular datasets.

Finding meaningful groups, i.e., clusters, in high-dimensional data such as images or texts without labeled data at hand is an important challenge in data mining. In recent years, deep clustering methods have achieved remarkable results in these tasks. However, most of these methods require the user to specify the number of clusters in advance. This is a major limitation since the number of clusters is typically unknown if labeled data is unavailable. Thus, an area of research has emerged that addresses this problem. Most of these approaches estimate the number of clusters separated from the clustering process. This results in a strong dependency of the clustering result on the quality of the initial embedding. Other approaches are tailored to specific clustering processes, making them hard to adapt to other scenarios. In this paper, we propose UNSEEN, a general framework that, starting from a given upper bound, is able to estimate the number of clusters. To the best of our knowledge, it is the first method that can be easily combined with various deep clustering algorithms. We demonstrate the applicability of our approach by combining UNSEEN with the popular deep clustering algorithms DCN, DEC, and DKM and verify its effectiveness through an extensive experimental evaluation on several image and tabular datasets. Moreover, we perform numerous ablations to analyze our approach and show the importance of its components.

AI-driven decision-making systems are becoming instrumental in the public sector, with applications spanning areas like criminal justice, social welfare, financial fraud detection, and public health. While these systems offer great potential benefits to institutional decision-making processes, such as improved efficiency and reliability, these systems face the challenge of aligning machine learning (ML) models with the complex realities of public sector decision-making. In this paper, we examine five key challenges where misalignment can occur, including distribution shifts, label bias, the influence of past decision-making on the data side, as well as competing objectives and human-in-the-loop on the model output side. Our findings suggest that standard ML methods often rely on assumptions that do not fully account for these complexities, potentially leading to unreliable and harmful predictions. To address this, we propose a shift in modeling efforts from focusing solely on predictive accuracy to improving decision-making outcomes. We offer guidance for selecting appropriate modeling frameworks, including counterfactual prediction and policy learning, by considering how the model estimand connects to the decision-maker’s utility. Additionally, we outline technical methods that address specific challenges within each modeling approach. Finally, we argue for the importance of external input from domain experts and stakeholders to ensure that model assumptions and design choices align with real-world policy objectives, taking a step towards harmonizing AI and public sector objectives.

Recent trends in Video Instance Segmentation (VIS) have seen a growing reliance on online methods to model complex and lengthy video sequences. However, the degradation of representation and noise accumulation of the online methods, especially during occlusion and abrupt changes, pose substantial challenges. Transformer-based query propagation provides promising directions at the cost of quadratic memory attention. However, they are susceptible to the degradation of instance features due to the above-mentioned challenges and suffer from cascading effects. The detection and rectification of such errors remain largely underexplored. To this end, we introduce textbf{GRAtt-VIS}, textbf{G}ated textbf{R}esidual textbf{Att}ention for textbf{V}ideo textbf{I}nstance textbf{S}egmentation. Firstly, we leverage a Gumbel-Softmax-based gate to detect possible errors in the current frame. Next, based on the gate activation, we rectify degraded features from its past representation. Such a residual configuration alleviates the need for dedicated memory and provides a continuous stream of relevant instance features. Secondly, we propose a novel inter-instance interaction using gate activation as a mask for self-attention. This masking strategy dynamically restricts the unrepresentative instance queries in the self-attention and preserves vital information for long-term tracking. We refer to this novel combination of Gated Residual Connection and Masked Self-Attention as textbf{GRAtt} block, which can easily be integrated into the existing propagation-based framework. Further, GRAtt blocks significantly reduce the attention overhead and simplify dynamic temporal modeling. GRAtt-VIS achieves state-of-the-art performance on YouTube-VIS and the highly challenging OVIS dataset, significantly improving over previous methods.

Earth Observation (EO) data analysis has been significantly revolutionized by deep learning (DL), with applications typically limited to grid-like data structures. Graph Neural Networks (GNNs) emerge as an important innovation, propelling DL into the non-Euclidean domain. Naturally, GNNs can effectively tackle the challenges posed by diverse modalities, multiple sensors, and the heterogeneous nature of EO data. To introduce GNNs in the related domains, our review begins by offering fundamental knowledge on GNNs. Then, we summarize the generic problems in EO, to which GNNs can offer potential solutions. Following this, we explore a broad spectrum of GNNs’ applications to scientific problems in Earth systems, covering areas such as weather and climate analysis, disaster management, air quality monitoring, agriculture, land cover classification, hydrological process modeling, and urban modeling. The rationale behind adopting GNNs in these fields is explained, alongside methodologies for organizing graphs and designing favorable architectures for various tasks. Furthermore, we highlight methodological challenges of implementing GNNs in these domains and possible solutions that could guide future research. While acknowledging that GNNs are not a universal solution, we conclude the paper by comparing them with other popular architectures like transformers and analyzing their potential synergies.

Global feature effect methods, such as partial dependence plots, provide an intelligible visualization of the expected marginal feature effect. However, such global feature effect methods can be misleading, as they do not represent local feature effects of single observations well when feature interactions are present. We formally introduce generalized additive decomposition of global effects (GADGET), which is a new framework based on recursive partitioning to find interpretable regions in the feature space such that the interaction-related heterogeneity of local feature effects is minimized. We provide a mathematical foundation of the framework and show that it is applicable to the most popular methods to visualize marginal feature effects, namely partial dependence, accumulated local effects, and Shapley additive explanations (SHAP) dependence. Furthermore, we introduce and validate a new permutation-based interaction test to detect significant feature interactions that is applicable to any feature effect method that fits into our proposed framework. We empirically evaluate the theoretical characteristics of the proposed methods based on various feature effect methods in different experimental settings. Moreover, we apply our introduced methodology to three real-world examples to showcase their usefulness.

Captions are a valuable scaffold for language learners, aiding comprehension and vocabulary acquisition. Past work has proposed enhancements such as keyword highlights for increased learning gains. However, little is known about learners’ experience with enhanced captions, although this is critical for adoption in everyday life. We conducted a survey and focus group to elicit learner preferences and requirements and implemented a processing pipeline for enhanced captions with keyword highlights, time-synchronized keyword highlights, and keyword captions. A subsequent online study (n = 66) showed that time-synchronized keyword highlights were the preferred design for learning but were perceived as too distracting to replace standard captions in everyday viewing scenarios. We conclude that keyword highlights and time-synchronization are suitable for integrating learning into an entertaining everyday- life activity, but the design should be optimized to provide a more seamless experience.

Topological correctness, i.e., the preservation of structural integrity and specific characteristics of shape, is a fundamental requirement for medical imaging tasks, such as neuron or vessel segmentation. Despite the recent surge in topology-aware methods addressing this challenge, their real-world applicability is hindered by flawed benchmarking practices. In this paper, we identify critical pitfalls in model evaluation that include inadequate connectivity choices, overlooked topological artifacts in ground truth annotations, and inappropriate use of evaluation metrics. Through detailed empirical analysis, we uncover these issues’ profound impact on the evaluation and ranking of segmentation methods. Drawing from our findings, we propose a set of actionable recommendations to establish fair and robust evaluation standards for topology-aware medical image segmentation methods.

Vision-language models (VLMs), such as CLIP and SigLIP, have found remarkable success in classification, retrieval, and generative tasks. For this, VLMs deterministically map images and text descriptions to a joint latent space in which their similarity is assessed using the cosine similarity. However, a deterministic mapping of inputs fails to capture uncertainties over concepts arising from domain shifts when used in downstream tasks. In this work, we propose post-hoc uncertainty estimation in VLMs that does not require additional training. Our method leverages a Bayesian posterior approximation over the last layers in VLMs and analytically quantifies uncertainties over cosine similarities. We demonstrate its effectiveness for uncertainty quantification and support set selection in active learning. Compared to baselines, we obtain improved and well-calibrated predictive uncertainties, interpretable uncertainty estimates, and sample-efficient active learning. Our results show promise for safety-critical applications of large-scale models.

The investigation of magnetic energy landscapes and the search for ground states of magnetic materials using ab initio methods like density functional theory (DFT) is a challenging task. Complex interactions, such as superexchange and spin-orbit coupling, make these calculations computationally expensive and often lead to non-trivial energy landscapes. Consequently, a comprehensive and systematic investigation of large magnetic configuration spaces is often impractical. We approach this problem by utilizing Bayesian Optimization, an active machine learning scheme that has proven to be efficient in modeling unknown functions and finding global minima. Using this approach we can obtain the magnetic contribution to the energy as a function of one or more spin canting angles with relatively small numbers of DFT calculations. To assess the capabilities and the efficiency of the approach we investigate the noncollinear magnetic energy landscapes of selected materials containing 3d, 5d and 5f magnetic ions: Ba3MnNb2O9, LaMn2Si2, β-MnO2, Sr2IrO4, UO2 and Ba2NaOsO6. By comparing our results to previous ab initio studies that followed more conventional approaches, we observe significant improvements in efficiency.

Disagreement in human labeling is ubiquitous, and can be captured in human judgment distributions (HJDs). Recent research has shown that explanations provide valuable information for understanding human label variation (HLV) and large language models (LLMs) can approximate HJD from a few human-provided label-explanation pairs. However, collecting explanations for every label is still time-consuming. This paper examines whether LLMs can be used to replace humans in generating explanations for approximating HJD. Specifically, we use LLMs as annotators to generate model explanations for a few given human labels. We test ways to obtain and combine these label-explanations with the goal to approximate human judgment distribution. We further compare the resulting human with model-generated explanations, and test automatic and human explanation selection. Our experiments show that LLM explanations are promising for NLI: to estimate HJD, generated explanations yield comparable results to human’s when provided with human labels. Importantly, our results generalize from datasets with human explanations to i) datasets where they are not available and ii) challenging out-of-distribution test sets.

Model merging combines multiple expert models - finetuned from a base foundation model on diverse tasks and domains - into a single, more capable model. However, most existing model merging approaches assume that all experts are available simultaneously. In reality, new tasks and domains emerge progressively over time, requiring strategies to integrate the knowledge of expert models as they become available: a process we call temporal model merging. The temporal dimension introduces unique challenges not addressed in prior work, raising new questions such as: when training for a new task, should the expert model start from the merged past experts or from the original base model? Should we merge all models at each time step? Which merging techniques are best suited for temporal merging? Should different strategies be used to initialize the training and deploy the model? To answer these questions, we propose a unified framework called TIME - Temporal Integration of Model Expertise - which defines temporal model merging across three axes: (1) Initialization Phase, (2) Deployment Phase, and (3) Merging Technique. Using TIME, we study temporal model merging across model sizes, compute budgets, and learning horizons on the FoMo-in-Flux benchmark. Our comprehensive suite of experiments across TIME allows us to uncover key insights for temporal model merging, offering a better understanding of current challenges and best practices for effective temporal model merging.

Digital pathology offers a groundbreaking opportunity to transform clinical practice in histopathological image analysis, yet faces a significant hurdle: the substantial file sizes of pathological Whole Slide Images (WSI). While current digital pathology solutions rely on lossy JPEG compression to address this issue, lossy compression can introduce color and texture disparities, potentially impacting clinical decision-making. While prior research addresses perceptual image quality and downstream performance independently of each other, we jointly evaluate compression schemes for perceptual and downstream task quality on four different datasets. In addition, we collect an initially uncompressed dataset for an unbiased perceptual evaluation of compression schemes. Our results show that deep learning models fine-tuned for perceptual quality outperform conventional compression schemes like JPEG-XL or WebP for further compression of WSI. However, they exhibit a significant bias towards the compression artifacts present in the training data and struggle to generalize across various compression schemes. We introduce a novel evaluation metric based on feature similarity between original files and compressed files that aligns very well with the actual downstream performance on the compressed WSI. Our metric allows for a general and standardized evaluation of lossy compression schemes and mitigates the requirement to independently assess different downstream tasks. Our study provides novel insights for the assessment of lossy compression schemes for WSI and encourages a unified evaluation of lossy compression schemes to accelerate the clinical uptake of digital pathology.

Feature-based explanations, using perturbations or gradients, are a prevalent tool to understand decisions of black box machine learning models. Yet, differences between these methods still remain mostly unknown, which limits their applicability for practitioners. In this work, we introduce a unified framework for local and global feature-based explanations using two well-established concepts: functional ANOVA (fANOVA) from statistics, and the notion of value and interaction from cooperative game theory. We introduce three fANOVA decompositions that determine the influence of feature distributions, and use game-theoretic measures, such as the Shapley value and interactions, to specify the influence of higher-order interactions. Our framework combines these two dimensions to uncover similarities and differences between a wide range of explanation techniques for features and groups of features. We then empirically showcase the usefulness of our framework on synthetic and real-world datasets.

Semantic correspondence, the task of determining relationships between different parts of images, underpins various applications including 3D reconstruction, image-to-image translation, object tracking, and visual place recognition. Recent studies have begun to explore representations learned in large generative image models for semantic correspondence, demonstrating promising results. Building on this progress, current state-of-the-art methods rely on combining multiple large models, resulting in high computational demands and reduced efficiency. In this work, we address this challenge by proposing a more computationally efficient approach. We propose a novel knowledge distillation technique to overcome the problem of reduced efficiency. We show how to use two large vision foundation models and distill the capabilities of these complementary models into one smaller model that maintains high accuracy at reduced computational cost. Furthermore, we demonstrate that by incorporating 3D data, we are able to further improve performance, without the need for human-annotated correspondences. Overall, our empirical results demonstrate that our distilled model with 3D data augmentation achieves performance superior to current state-of-the-art methods while significantly reducing computational load and enhancing practicality for real-world applications, such as semantic video correspondence. Our code and weights are publicly available on our project page.

Making inferences about physical properties of the Universe requires knowledge of the data likelihood. A Gaussian distribution is commonly assumed for the uncertainties with a covariance matrix estimated from a set of simulations. The noise in such covariance estimates causes two problems: it distorts the width of the parameter contours, and it adds scatter to the location of those contours which is not captured by the widths themselves. For non-Gaussian likelihoods, an approximation may be derived via Simulation-Based Inference (SBI). It is often implicitly assumed that parameter constraints from SBI analyses, which do not use covariance matrices, are not affected by the same problems as parameter estimation with a covariance matrix estimated from simulations. We investigate whether SBI suffers from effects similar to those of covariance estimation in Gaussian likelihoods. We use Neural Posterior and Likelihood Estimation with continuous and masked autoregressive normalizing flows for density estimation. We fit our approximate posterior models to simulations drawn from a Gaussian linear model, so that the SBI result can be compared to the true posterior. We test linear and neural network based compression, demonstrating that neither methods circumvent the issues of covariance estimation. SBI suffers an inflation of posterior variance that is equal or greater than the analytical result in covariance estimation for Gaussian likelihoods for the same number of simulations. The assumption that SBI requires a smaller number of simulations than covariance estimation for a Gaussian likelihood analysis is inaccurate. The limitations of traditional likelihood analysis with simulation-based covariance remain for SBI with a finite simulation budget. Despite these issues, we show that SBI correctly draws the true posterior contour given enough simulations.

In generative models, two paradigms have gained attraction in various applications: next-set prediction-based Masked Generative Models and next-noise prediction-based Non-Autoregressive Models, e.g., Diffusion Models. In this work, we propose using discrete-state models to connect them and explore their scalability in the vision domain. First, we conduct a step-by-step analysis in a unified design space across two types of models including timestep-independence, noise schedule, temperature, guidance strength, etc in a scalable manner. Second, we re-cast typical discriminative tasks, e.g., image segmentation, as an unmasking process from [MASK] tokens on a discrete-state model. This enables us to perform various sampling processes, including flexible conditional sampling by only training once to model the joint distribution. All aforementioned explorations lead to our framework named Discrete Interpolants, which enables us to achieve state-of-the-art or competitive performance compared to previous discrete-state based methods in various benchmarks, like ImageNet256, MS COCO, and video dataset FaceForensics. In summary, by leveraging [MASK] in discrete-state models, we can bridge Masked Generative and Non-autoregressive Diffusion models, as well as generative and discriminative tasks.

Vision-Language Models (VLMs) trained with contrastive loss have achieved significant advancements in various vision and language tasks. However, the global nature of contrastive loss makes VLMs focus predominantly on foreground objects, neglecting other crucial information in the image, which limits their effectiveness in downstream tasks. To address these challenges, we propose COSMOS: CrOSs-MOdality Self-distillation for vision-language pre-training that integrates a novel text-cropping strategy and cross-attention module into a self-supervised learning framework. We create global and local views of images and texts (i.e., multi-modal augmentations), which are essential for self-distillation in VLMs. We further introduce a cross-attention module, enabling COSMOS to learn comprehensive cross-modal representations optimized via a cross-modality self-distillation loss. COSMOS consistently outperforms previous strong baselines on various zero-shot downstream tasks, including retrieval, classification, and semantic segmentation. Additionally, it surpasses CLIP-based models trained on larger datasets in visual perception and contextual understanding tasks.

The problem of symbolic regression (SR) arises in many different applications, such as identifying physical laws or deriving mathematical equations describing the behavior of financial markets from given data. Various methods exist to address the problem of SR, often based on genetic programming. However, these methods are usually complicated and involve various hyperparameters. In this paper, we present our new approach ParFam that utilizes parametric families of suitable symbolic functions to translate the discrete symbolic regression problem into a continuous one, resulting in a more straightforward setup compared to current state-of-the-art methods. In combination with a global optimizer, this approach results in a highly effective method to tackle the problem of SR. We theoretically analyze the expressivity of ParFam and demonstrate its performance with extensive numerical experiments based on the common SR benchmark suit SRBench, showing that we achieve state-of-the-art results. Moreover, we present an extension incorporating a pre-trained transformer network DL-ParFam to guide ParFam, accelerating the optimization process by up to two magnitudes.

3D Gaussian Splatting has recently gained traction for its efficient training and real-time rendering. While the vanilla Gaussian Splatting representation is mainly designed for view synthesis, more recent works investigated how to extend it with scene understanding and language features. However, existing methods lack a detailed comprehension of scenes, limiting their ability to segment and interpret complex structures. To this end, We introduce SuperGSeg, a novel approach that fosters cohesive, context-aware scene representation by disentangling segmentation and language field distillation. SuperGSeg first employs neural Gaussians to learn instance and hierarchical segmentation features from multi-view images with the aid of off-the-shelf 2D masks. These features are then leveraged to create a sparse set of what we call Super-Gaussians. Super-Gaussians facilitate the distillation of 2D language features into 3D space. Through Super-Gaussians, our method enables high-dimensional language feature rendering without extreme increases in GPU memory. Extensive experiments demonstrate that SuperGSeg outperforms prior works on both open-vocabulary object localization and semantic segmentation tasks.

In recent research, large language models (LLMs) have been increasingly used to investigate public opinions. This study investigates the algorithmic fidelity of LLMs, i.e., the ability to replicate the socio-cultural context and nuanced opinions of human participants. Using open-ended survey data from the German Longitudinal Election Studies (GLES), we prompt different LLMs to generate synthetic public opinions reflective of German subpopulations by incorporating demographic features into the persona prompts. Our results show that Llama performs better than other LLMs at representing subpopulations, particularly when there is lower opinion diversity within those groups. Our findings further reveal that the LLM performs better for supporters of left-leaning parties like The Greens and The Left compared to other parties, and matches the least with the right-party AfD. Additionally, the inclusion or exclusion of specific variables in the prompts can significantly impact the models’ predictions. These findings underscore the importance of aligning LLMs to more effectively model diverse public opinions while minimizing political biases and enhancing robustness in representativeness.

Accurately describing images with text is a foundation of explainable AI. Vision-Language Models (VLMs) like CLIP have recently addressed this by aligning images and texts in a shared embedding space, expressing semantic similarities between vision and language embeddings. VLM classification can be improved with descriptions generated by Large Language Models (LLMs). However, it is difficult to determine the contribution of actual description semantics, as the performance gain may also stem from a semantic-agnostic ensembling effect, where multiple modified text prompts act as a noisy test-time augmentation for the original one. We propose an alternative evaluation scenario to decide if a performance boost of LLM-generated descriptions is caused by such a noise augmentation effect or rather by genuine description semantics. The proposed scenario avoids noisy test-time augmentation and ensures that genuine, distinctive descriptions cause the performance boost. Furthermore, we propose a training-free method for selecting discriminative descriptions that work independently of classname-ensembling effects. Our approach identifies descriptions that effectively differentiate classes within a local CLIP label neighborhood, improving classification accuracy across seven datasets. Additionally, we provide insights into the explainability of description-based image classification with VLMs.

We investigate biases in pretraining datasets for large language models (LLMs) through dataset classification experiments. Building on prior work demonstrating the existence of biases in popular computer vision datasets, we analyze popular open-source pretraining datasets for LLMs derived from CommonCrawl including C4, RefinedWeb, DolmaCC, RedPajama-V2, FineWeb, and DCLM-Baseline. Despite those datasets being obtained with similar filtering and deduplication steps, neural networks can classify surprisingly well which dataset a single text sequence belongs to, significantly better than a human can. This indicates that popular pretraining datasets have their own unique biases or fingerprints. Those biases remain even when the text is rewritten with LLMs. Moreover, these biases propagate through training: Random sequences generated by models trained on those datasets can be classified well by a classifier trained on the original datasets.

Transforming CO2 into methanol represents a crucial step towards closing the carbon cycle, with thermoreduction technology nearing industrial application. However, obtaining high methanol yields and ensuring the stability of heterocatalysts remain significant challenges. Herein, we present a sophisticated computational framework to accelerate the discovery of novel thermal heterogeneous catalysts, using machine-learned force fields. We propose a new catalytic descriptor, termed adsorption energy distribution, that aggregates the binding energies for different catalyst facets, binding sites, and adsorbates. The descriptor is versatile and can easily be adjusted to a specific reaction through careful choice of the key-step reactants and reaction intermediates. By applying unsupervised machine learning and statistical analysis to a dataset comprising nearly 160 metallic alloys, we offer a powerful tool for catalyst discovery. Finally, we propose new promising candidate materials such as ZnRh and ZnPt3, which to our knowledge, have not yet been tested, and discuss their possible advantage in terms of stability.

Image registration is fundamental in medical imaging applications, such as disease progression analysis or radiation therapy planning. The primary objective of image registration is to precisely capture the deformation between two or more images, typically achieved by minimizing an optimization problem. Due to its inherent ill-posedness, regularization is a key component in driving the solution toward anatomically meaningful deformations. A wide range of regularization methods has been proposed for both conventional and deep learning-based registration. However, the appropriate application of regularization techniques often depends on the specific registration problem, and no one-fits-all method exists. Despite its importance, regularization is often overlooked or addressed with default approaches, assuming existing methods are sufficient. A comprehensive and structured review remains missing. This review addresses this gap by introducing a novel taxonomy that systematically categorizes the diverse range of proposed regularization methods. It highlights the emerging field of learned regularization, which leverages data-driven techniques to automatically derive deformation properties from the data. Moreover, this review examines the transfer of regularization methods from conventional to learning-based registration, identifies open challenges, and outlines future research directions. By emphasizing the critical role of regularization in image registration, we hope to inspire the research community to reconsider regularization strategies in modern registration algorithms and to explore this rapidly evolving field further.

Machine learning models are highly vulnerable to label flipping, i.e., the adversarial modification (poisoning) of training labels to compromise performance. Thus, deriving robustness certificates is important to guarantee that test predictions remain unaffected and to understand worst-case robustness behavior. However, for Graph Neural Networks (GNNs), the problem of certifying label flipping has so far been unsolved. We change this by introducing an exact certification method, deriving both sample-wise and collective certificates. Our method leverages the Neural Tangent Kernel (NTK) to capture the training dynamics of wide networks enabling us to reformulate the bilevel optimization problem representing label flipping into a Mixed-Integer Linear Program (MILP). We apply our method to certify a broad range of GNN architectures in node classification tasks. Thereby, concerning the worst-case robustness to label flipping: (i) we establish hierarchies of GNNs on different benchmark graphs; (ii) quantify the effect of architectural choices such as activations, depth and skip-connections; and surprisingly, (iii) uncover a novel phenomenon of the robustness plateauing for intermediate perturbation budgets across all investigated datasets and architectures. While we focus on GNNs, our certificates are applicable to sufficiently wide NNs in general through their NTK. Thus, our work presents the first exact certificate to a poisoning attack ever derived for neural networks, which could be of independent interest.

Adequately generating and evaluating prediction models based on supervised machine learning (ML) is often challenging, especially for less experienced users in applied research areas. Special attention is required in settings where the model generation process involves hyperparameter tuning, i.e. data-driven optimization of different types of hyperparameters to improve the predictive performance of the resulting model. Discussions about tuning typically focus on the hyperparameters of the ML algorithm (e.g., the minimum number of observations in each terminal node for a tree-based algorithm). In this context, it is often neglected that hyperparameters also exist for the preprocessing steps that are applied to the data before it is provided to the algorithm (e.g., how to handle missing feature values in the data). As a consequence, users experimenting with different preprocessing options to improve model performance may be unaware that this constitutes a form of hyperparameter tuning - albeit informal and unsystematic - and thus may fail to report or account for this optimization. To illuminate this issue, this paper reviews and empirically illustrates different procedures for generating and evaluating prediction models, explicitly addressing the different ways algorithm and preprocessing hyperparameters are typically handled by applied ML users. By highlighting potential pitfalls, especially those that may lead to exaggerated performance claims, this review aims to further improve the quality of predictive modeling in ML applications.

Internal features from large-scale pre-trained diffusion models have recently been established as powerful semantic descriptors for a wide range of downstream tasks. Works that use these features generally need to add noise to images before passing them through the model to obtain the semantic features, as the models do not offer the most useful features when given images with little to no noise. We show that this noise has a critical impact on the usefulness of these features that cannot be remedied by ensembling with different random noises. We address this issue by introducing a lightweight, unsupervised fine-tuning method that enables diffusion backbones to provide high-quality, noise-free semantic features. We show that these features readily outperform previous diffusion features by a wide margin in a wide variety of extraction setups and downstream tasks, offering better performance than even ensemble-based methods at a fraction of the cost.

Ambiguity resolution is key to effective communication. While humans effortlessly address ambiguity through conversational grounding strategies, the extent to which current language models can emulate these strategies remains unclear. In this work, we examine referential ambiguity in image-based question answering by introducing RACQUET, a carefully curated dataset targeting distinct aspects of ambiguity. Through a series of evaluations, we reveal significant limitations and problems of overconfidence of state-of-the-art large multimodal language models in addressing ambiguity in their responses. The overconfidence issue becomes particularly relevant for RACQUET-BIAS, a subset designed to analyze a critical yet underexplored problem: failing to address ambiguity leads to stereotypical, socially biased responses. Our results underscore the urgency of equipping models with robust strategies to deal with uncertainty without resorting to undesirable stereotypes.

Vision tokenizers have gained a lot of attraction due to their scalability and compactness; previous works depend on old-school GAN-based hyperparameters, biased comparisons, and a lack of comprehensive analysis of the scaling behaviours. To tackle those issues, we introduce Grouped Spherical Quantization (GSQ), featuring spherical codebook initialization and lookup regularization to constrain codebook latent to a spherical surface. Our empirical analysis of image tokenizer training strategies demonstrates that GSQ-GAN achieves superior reconstruction quality over state-of-the-art methods with fewer training iterations, providing a solid foundation for scaling studies. Building on this, we systematically examine the scaling behaviours of GSQ, specifically in latent dimensionality, codebook size, and compression ratios, and their impact on model performance. Our findings reveal distinct behaviours at high and low spatial compression levels, underscoring challenges in representing high-dimensional latent spaces. We show that GSQ can restructure high-dimensional latent into compact, low-dimensional spaces, thus enabling efficient scaling with improved quality. As a result, GSQ-GAN achieves a 16x down-sampling with a reconstruction FID (rFID) of 0.50.

Uncertainty quantification (UQ) is essential for assessing the reliability of Earth observation (EO) products. However, the extensive use of machine learning models in EO introduces an additional layer of complexity, as those models themselves are inherently uncertain. While various UQ methods do exist for machine learning models, their performance on EO datasets remains largely unevaluated. A key challenge in the community is the absence of the ground truth for uncertainty, i.e. how certain the uncertainty estimates are, apart from the labels for the image/signal. This article fills this gap by introducing three benchmark datasets specifically designed for UQ in EO machine learning models. These datasets address three common problem types in EO: regression, image segmentation, and scene classification. They enable a transparent comparison of different UQ methods for EO machine learning models. We describe the creation and characteristics of each dataset, including data sources, preprocessing steps, and label generation, with a particular focus on calculating the reference uncertainty. We also showcase baseline performance of several machine learning models on each dataset, highlighting the utility of these benchmarks for model development and comparison. Overall, this article offers a valuable resource for researchers and practitioners working in artificial intelligence for EO, promoting a more accurate and reliable quality measure of the outputs of machine learning models.

Constructing confidence intervals (CIs) for the average treatment effect (ATE) from patient records is crucial to assess the effectiveness and safety of drugs. However, patient records typically come from different hospitals, thus raising the question of how multiple observational datasets can be effectively combined for this purpose. In our paper, we propose a new method that estimates the ATE from multiple observational datasets and provides valid CIs. Our method makes little assumptions about the observational datasets and is thus widely applicable in medical practice. The key idea of our method is that we leverage prediction-powered inferences and thereby essentially `shrink’ the CIs so that we offer more precise uncertainty quantification as compared to naïve approaches. We further prove the unbiasedness of our method and the validity of our CIs. We confirm our theoretical results through various numerical experiments. Finally, we provide an extension of our method for constructing CIs from combinations of experimental and observational datasets.

Biophysical modeling of brain tumors has emerged as a promising strategy for personalizing radiotherapy planning by estimating the otherwise hidden distribution of tumor cells within the brain. However, many existing state-of-the-art methods are computationally intensive, limiting their widespread translation into clinical practice. In this work, we propose an efficient and direct method that utilizes soft physical constraints to estimate the tumor cell concentration from preoperative MRI of brain tumor patients. Our approach optimizes a 3D tumor concentration field by simultaneously minimizing the difference between the observed MRI and a physically informed loss function. Compared to existing state-of-the-art techniques, our method significantly improves predicting tumor recurrence on two public datasets with a total of 192 patients while maintaining a clinically viable runtime of under one minute - a substantial reduction from the 30 minutes required by the current best approach. Furthermore, we showcase the generalizability of our framework by incorporating additional imaging information and physical constraints, highlighting its potential to translate to various medical diffusion phenomena with imperfect data.

To date, most place recognition methods focus on single-modality retrieval. While they perform well in specific environments, cross-modal methods offer greater flexibility by allowing seamless switching between map and query sources. It also promises to reduce computation requirements by having a unified model, and achieving greater sample efficiency by sharing parameters. In this work, we develop a universal solution to place recognition, UniLoc, that works with any single query modality (natural language, image, or point cloud). UniLoc leverages recent advances in large-scale contrastive learning, and learns by matching hierarchically at two levels: instance-level matching and scene-level matching. Specifically, we propose a novel Self-Attention based Pooling (SAP) module to evaluate the importance of instance descriptors when aggregated into a place-level descriptor. Experiments on the KITTI-360 dataset demonstrate the benefits of cross-modality for place recognition, achieving superior performance in cross-modal settings and competitive results also for uni-modal scenarios.

We tackle the problem of localizing the traffic surveillance cameras in cooperative perception. To overcome the lack of large-scale real-world intersection datasets, we introduce Carla Intersection, a new simulated dataset with 75 urban and rural intersections in Carla. Moreover, we introduce a novel neural network, TrafficLoc, localizing traffic cameras within a 3D reference map. TrafficLoc employs a coarse-to-fine matching pipeline. For image-point cloud feature fusion, we propose a novel Geometry-guided Attention Loss to address cross-modal viewpoint inconsistencies. During coarse matching, we propose an Inter-Intra Contrastive Learning to achieve precise alignment while preserving distinctiveness among local intra-features within image patch-point group pairs. Besides, we introduce Dense Training Alignment with a soft-argmax operator to consider additional features when regressing the final position. Extensive experiments show that our TrafficLoc improves the localization accuracy over the state-of-the-art Image-to-point cloud registration methods by a large margin (up to 86%) on Carla Intersection and generalizes well to real-world data. TrafficLoc also achieves new SOTA performance on KITTI and NuScenes datasets, demonstrating strong localization ability across both in-vehicle and traffic cameras.

CLIP has shown impressive results in aligning images and texts at scale. However, its ability to capture detailed visual features remains limited because CLIP matches images and texts at a global level. To address this issue, we propose FLAIR, Fine-grained Language-informed Image Representations, an approach that utilizes long and detailed image descriptions to learn localized image embeddings. By sampling diverse sub-captions that describe fine-grained details about an image, we train our vision-language model to produce not only global embeddings but also text-specific image representations. Our model introduces text-conditioned attention pooling on top of local image tokens to produce fine-grained image representations that excel at retrieving detailed image content. We achieve state-of-the-art performance on both, existing multimodal retrieval benchmarks, as well as, our newly introduced fine-grained retrieval task which evaluates vision-language models’ ability to retrieve partial image content. Furthermore, our experiments demonstrate the effectiveness of FLAIR trained on 30M image-text pairs in capturing fine-grained visual information, including zero-shot semantic segmentation, outperforming models trained on billions of pairs.

The rapid evolution of Vision Language Models (VLMs) has catalyzed significant advancements in artificial intelligence, expanding research across various disciplines, including Earth Observation (EO). While VLMs have enhanced image understanding and data processing within EO, their applications have predominantly focused on image content description. This limited focus overlooks their potential in geographic and scientific regression tasks, which are essential for diverse EO applications. To bridge this gap, this paper introduces a novel benchmark dataset, called textbf{REO-Instruct} to unify regression and generation tasks specifically for the EO domain. Comprising 1.6 million multimodal EO imagery and language pairs, this dataset is designed to support both biomass regression and image content interpretation tasks. Leveraging this dataset, we develop REO-VLM, a groundbreaking model that seamlessly integrates regression capabilities with traditional generative functions. By utilizing language-driven reasoning to incorporate scientific domain knowledge, REO-VLM goes beyond solely relying on EO imagery, enabling comprehensive interpretation of complex scientific attributes from EO data. This approach establishes new performance benchmarks and significantly enhances the capabilities of environmental monitoring and resource management.

Hate speech online remains an understudied issue for marginalized communities, and has seen rising relevance, especially in the Global South, which includes developing societies with increasing internet penetration. In this paper, we aim to provide marginalized communities living in societies where the dominant language is low-resource with a privacy-preserving tool to protect themselves from hate speech on the internet by filtering offensive content in their native languages. Our contribution in this paper is twofold: 1) we release REACT (REsponsive hate speech datasets Across ConTexts), a collection of high-quality, culture-specific hate speech detection datasets comprising seven distinct target groups in eight low-resource languages, curated by experienced data collectors; 2) we propose a solution to few-shot hate speech detection utilizing federated learning (FL), a privacy-preserving and collaborative learning approach, to continuously improve a central model that exhibits robustness when tackling different target groups and languages. By keeping the training local to the users’ devices, we ensure the privacy of the users’ data while benefitting from the efficiency of federated learning. Furthermore, we personalize client models to target-specific training data and evaluate their performance. Our results indicate the effectiveness of FL across different target groups, whereas the benefits of personalization on few-shot learning are not clear.

The interest in matching non-rigidly deformed shapes represented as raw point clouds is rising due to the proliferation of low-cost 3D sensors. Yet, the task is challenging since point clouds are irregular and there is a lack of intrinsic shape information. We propose to tackle these challenges by learning a new shape representation – a per-point high dimensional embedding, in an embedding space where semantically similar points share similar embeddings. The learned embedding has multiple beneficial properties: it is aware of the underlying shape geometry and is robust to shape deformations and various shape artefacts, such as noise and partiality. Consequently, this embedding can be directly employed to retrieve high-quality dense correspondences through a simple nearest neighbor search in the embedding space. Extensive experiments demonstrate new state-of-the-art results and robustness in numerous challenging non-rigid shape matching benchmarks and show its great potential in other shape analysis tasks, such as segmentation.

The role of subword segmentation in relation to capturing morphological patterns in LLMs is currently not well explored. Ideally, one would train models like GPT using various segmentations and evaluate how well word meanings are captured. Since this is not computationally feasible, we group words according to their segmentation properties and compare how well a model can solve a linguistic task for these groups. We study two criteria: (i) adherence to morpheme boundaries and (ii) the segmentation consistency of the different inflected forms of a lemma. We select word forms with high and low values for these criteria and carry out experiments on GPT-4o’s ability to capture verbal inflection for 10 languages. Our results indicate that in particular the criterion of segmentation consistency can help to predict the model’s ability to recognize and generate the lemma from an inflected form, providing evidence that subword segmentation is relevant.

Large Language Models (LLMs) show remarkable performance on a wide variety of tasks. Most LLMs split text into multi-character tokens and process them as atomic units without direct access to individual characters. This raises the question: To what extent can LLMs learn orthographic information? To answer this, we propose a new benchmark, CUTE, which features a collection of tasks designed to test the orthographic knowledge of LLMs. We evaluate popular LLMs on CUTE, finding that most of them seem to know the spelling of their tokens, yet fail to use this information effectively to manipulate text, calling into question how much of this knowledge is generalizable.

Text style transfer (TST) aims to modify the style of a text without altering its original meaning. Large language models (LLMs) demonstrate superior performance across multiple tasks, including TST. However, in zero-shot setups, they tend to directly copy a significant portion of the input text to the output without effectively changing its style. To enhance the stylistic variety and fluency of the text, we present sNeuron-TST, a novel approach for steering LLMs using style-specific neurons in TST. Specifically, we identify neurons associated with the source and target styles and deactivate source-style-only neurons to give target-style words a higher probability, aiming to enhance the stylistic diversity of the generated text. However, we find that this deactivation negatively impacts the fluency of the generated text, which we address by proposing an improved contrastive decoding method that accounts for rapid token probability shifts across layers caused by deactivated source-style neurons. Empirical experiments demonstrate the effectiveness of the proposed method on six benchmarks, encompassing formality, toxicity, politics, politeness, authorship, and sentiment.

Full-parameter fine-tuning has become the go-to choice for adapting language models (LMs) to downstream tasks due to its excellent performance. As LMs grow in size, fine-tuning the full parameters of LMs requires a prohibitively large amount of GPU memory. Existing approaches utilize zeroth-order optimizer to conserve GPU memory, which can potentially compromise the performance of LMs as non-zero order optimizers tend to converge more readily on most downstream tasks. In this paper, we propose a novel optimizer-independent end-to-end hierarchical fine-tuning strategy, HiFT, which only updates a subset of parameters at each training step. HiFT can significantly reduce the amount of gradients and optimizer state parameters residing in GPU memory at the same time, thereby reducing GPU memory usage. Our results demonstrate that: (1) HiFT achieves comparable performance to parameter-efficient fine-tuning and standard full parameter fine-tuning. (2) HiFT supports various optimizers including AdamW, AdaGrad, SGD, etc. (3) HiFT can save more than 60% GPU memory compared with standard full-parameter fine-tuning for 7B model. (4) HiFT enables full-parameter fine-tuning of a 7B model on single 48G A6000 with a precision of 32 using the AdamW optimizer, without using any memory saving techniques.

Knights and knaves problems represent a classic genre of logical puzzles where characters either tell the truth or lie. The objective is to logically deduce each character’s identity based on their statements. The challenge arises from the truth-telling or lying behavior, which influences the logical implications of each statement. Solving these puzzles requires not only direct deductions from individual statements, but the ability to assess the truthfulness of statements by reasoning through various hypothetical scenarios. As such, knights and knaves puzzles serve as compelling examples of suppositional reasoning. In this paper, we introduce TruthQuest, a benchmark for suppositional reasoning based on the principles of knights and knaves puzzles. Our benchmark presents problems of varying complexity, considering both the number of characters and the types of logical statements involved. Evaluations on TruthQuest show that large language models like Llama 3 and Mixtral-8x7B exhibit significant difficulties solving these tasks. A detailed error analysis of the models’ output reveals that lower-performing models exhibit a diverse range of reasoning errors, frequently failing to grasp the concept of truth and lies. In comparison, more proficient models primarily struggle with accurately inferring the logical implications of potentially false statements.

Human label variation (HLV) is a valuable source of information that arises when multiple human annotators provide different labels for valid reasons. In Natural Language Inference (NLI) earlier approaches to capturing HLV involve either collecting annotations from many crowd workers to represent human judgment distribution (HJD) or use expert linguists to provide detailed explanations for their chosen labels. While the former method provides denser HJD information, obtaining it is resource-intensive. In contrast, the latter offers richer textual information but it is challenging to scale up to many human judges. Besides, large language models (LLMs) are increasingly used as evaluators (‘LLM judges’) but with mixed results, and few works aim to study HJDs. This study proposes to exploit LLMs to approximate HJDs using a small number of expert labels and explanations. Our experiments show that a few explanations significantly improve LLMs’ ability to approximate HJDs with and without explicit labels, thereby providing a solution to scale up annotations for HJD. However, fine-tuning smaller soft-label aware models with the LLM-generated model judgment distributions (MJDs) presents partially inconsistent results: while similar in distance, their resulting fine-tuned models and visualized distributions differ substantially. We show the importance of complementing instance-level distance measures with a global-level shape metric and visualization to more effectively evaluate MJDs against human judgment distributions.

Stemming from traditional knowledge graphs (KGs), hyper-relational KGs (HKGs) provide additional key-value pairs (i.e., qualifiers) for each KG fact that help to better restrict the fact validity. In recent years, there has been an increasing interest in studying graph reasoning over HKGs. Meanwhile, as discussed in recent works that focus on temporal KGs (TKGs), world knowledge is ever-evolving, making it important to reason over temporal facts in KGs. Previous mainstream benchmark HKGs do not explicitly specify temporal information for each HKG fact. Therefore, almost all existing HKG reasoning approaches do not devise any module specifically for temporal reasoning. To better study temporal fact reasoning over HKGs, we propose a new type of data structure named hyper-relational TKG (HTKG). Every fact in an HTKG is coupled with a timestamp explicitly indicating its time validity. We develop two new benchmark HTKG datasets, i.e., Wiki-hy and YAGO-hy, and propose an HTKG reasoning model that efficiently models hyper-relational temporal facts. To support future research on this topic, we open-source our datasets and model.

Decoding from the output distributions of large language models to produce high-quality text is a complex challenge in language modeling. Various approaches, such as beam search, sampling with temperature, k−sampling, nucleus p−sampling, typical decoding, contrastive decoding, and contrastive search, have been proposed to address this problem, aiming to improve coherence, diversity, as well as resemblance to human-generated text. In this study, we introduce adaptive contrastive search, a novel decoding strategy extending contrastive search by incorporating an adaptive degeneration penalty, guided by the estimated uncertainty of the model at each generation step. This strategy is designed to enhance both the creativity and diversity of the language modeling process while at the same time producing coherent and high-quality generated text output. Our findings indicate performance enhancement in both aspects, across different model architectures and datasets, underscoring the effectiveness of our method in text generation tasks. Our code base, datasets, and models are publicly available.

Instruction tuning enables language models to more effectively generalize and better follow user intent. However, obtaining instruction data is costly and challenging. Prior work employs methods such as expensive human annotation, crowd-sourced datasets with alignment issues, and generating noisy examples via LLMs. We introduce the LongForm-C dataset, which is created by reverse instructions. We generate instructions via LLMs for human-written corpus examples using reverse instructions. First we select a diverse set of human-written documents from corpora such as C4 and Wikipedia; then we generate instructions for these documents via LLMs. This approach provides a cheaper and cleaner instruction-tuning dataset with natural output and one suitable for long text generation. Our models outperform 10x larger language models without instruction tuning on tasks such as story/recipe generation and long-form question answering. Moreover, LongForm models outperform prior instruction-tuned models such as FLAN-T5 and Alpaca by a large margin, and improve language understanding capabilities further.

In the video-language domain, recent works in leveraging zero-shot Large Language Model-based reasoning for video understanding have become competitive challengers to previous end-to-end models. However, long video understanding presents unique challenges due to the complexity of reasoning over extended timespans, even for zero-shot LLM-based approaches. The challenge of information redundancy in long videos prompts the question of what specific information is essential for large language models (LLMs) and how to leverage them for complex spatial-temporal reasoning in long-form video analysis. We propose a framework VideoINSTA, i.e. INformative Spatial-TemporAl Reasoning for zero-shot long-form video understanding. VideoINSTA contributes (1) a zero-shot framework for long video understanding using LLMs; (2) an event-based temporal reasoning and content-based spatial reasoning approach for LLMs to reason over spatial-temporal information in videos; (3) a self-reflective information reasoning scheme balancing temporal factors based on information sufficiency and prediction confidence. Our model significantly improves the state-of-the-art on three long video question-answering benchmarks: EgoSchema, NextQA, and IntentQA, and the open question answering dataset ActivityNetQA.

Many datasets have been developed to train and evaluate document-level relation extraction (RE) models. Most of these are constructed using real-world data. It has been shown that RE models trained on real-world data suffer from factual biases. To evaluate and address this issue, we present CovEReD, a counterfactual data generation approach for document-level relation extraction datasets using entity replacement. We first demonstrate that models trained on factual data exhibit inconsistent behavior: while they accurately extract triples from factual data, they fail to extract the same triples after counterfactual modification. This inconsistency suggests that models trained on factual data rely on spurious signals such as specific entities and external knowledge – rather than on the input context – to extract triples. We show that by generating document-level counterfactual data with CovEReD and training models on them, consistency is maintained with minimal impact on RE performance. We release our CovEReD pipeline as well as Re-DocRED-CF, a dataset of counterfactual RE documents, to assist in evaluating and addressing inconsistency in document-level RE.

One of the major aspects contributing to the striking performance of large language models (LLMs) is the vast amount of factual knowledge accumulated during pre-training. Yet, many LLMs suffer from self-inconsistency, which raises doubts about their trustworthiness and reliability. This paper focuses on entity type ambiguity, analyzing the proficiency and consistency of state-of-the-art LLMs in applying factual knowledge when prompted with ambiguous entities. To do so, we propose an evaluation protocol that disentangles knowing from applying knowledge, and test state-of-the-art LLMs on 49 ambiguous entities. Our experiments reveal that LLMs struggle with choosing the correct entity reading, achieving an average accuracy of only 85%, and as low as 75% with underspecified prompts. The results also reveal systematic discrepancies in LLM behavior, showing that while the models may possess knowledge, they struggle to apply it consistently, exhibit biases toward preferred readings, and display self-inconsistencies. This highlights the need to address entity ambiguity in the future for more trustworthy LLMs.

To ensure large language models contain up-to-date knowledge, they need to be updated regularly. However, model editing is challenging as it might also affect knowledge that is unrelated to the new data. State-of-the-art methods identify parameters associated with specific knowledge and then modify them via direct weight updates. However, these locate-and-edit methods suffer from heavy computational overhead and lack theoretical validation. In contrast, directly fine-tuning the model on requested edits affects the model’s behavior on unrelated knowledge, and significantly damages the model’s generation fluency and consistency. To address these challenges, we propose SAUL, a streamlined model editing method that uses sentence concatenation with augmented random facts for generation regularization. Evaluations on three model editing benchmarks show that SAUL is a practical and reliable solution for model editing outperforming state-of-the-art methods while maintaining generation quality and reducing computational overhead.

Multilingual pre-trained models (mPLMs) have shown impressive performance on cross-lingual transfer tasks. However, the transfer performance is often hindered when a low-resource target language is written in a different script than the high-resource source language, even though the two languages may be related or share parts of their vocabularies. Inspired by recent work that uses transliteration to address this problem, our paper proposes a transliteration-based post-pretraining alignment (PPA) method aiming to improve the cross-lingual alignment between languages using diverse scripts. We select two areal language groups, Mediterranean-Amharic-Farsi and South+East Asian Languages, wherein the languages are mutually influenced but use different scripts. We apply our method to these language groups and conduct extensive experiments on a spectrum of downstream tasks. The results show that after PPA, models consistently outperform the original model (up to 50% for some tasks) in English-centric transfer. In addition, when we use languages other than English as sources in transfer, our method obtains even larger improvements.

Multiple choice question answering tasks evaluate the reasoning, comprehension, and mathematical abilities of Large Language Models (LLMs). While existing benchmarks employ automatic translation for multilingual evaluation, this approach is error-prone and potentially introduces culturally biased questions, especially in social sciences. We introduce the first multitask, multiple-choice Turkish QA benchmark, TurkishMMLU, to evaluate LLMs’ understanding of the Turkish language. TurkishMMLU includes over 10,000 questions, covering 9 different subjects from Turkish high-school education curricula. These questions are written by curriculum experts, suitable for the high-school curricula in Turkey, covering subjects ranging from natural sciences and math questions to more culturally representative topics such as Turkish Literature and the history of the Turkish Republic. We evaluate over 20 LLMs, including multilingual open-source (e.g., Gemma, Llama, MT5), closed-source (GPT 4o, Claude, Gemini), and Turkish-adapted (e.g., Trendyol) models. We provide an extensive evaluation, including zero-shot and few-shot evaluation of LLMs, chain-of-thought reasoning, and question difficulty analysis along with model performance. We provide an in-depth analysis of the Turkish capabilities and limitations of current LLMs to provide insights for future LLMs for the Turkish language.

Question decomposition has emerged as an effective strategy for prompting Large Language Models (LLMs) to answer complex questions. However, while existing methods primarily focus on unimodal language models, the question decomposition capability of Multimodal Large Language Models (MLLMs) has yet to be explored. To this end, this paper explores visual question decomposition on MLLMs. Specifically, we introduce a systematic evaluation framework including a dataset and several evaluation criteria to assess the quality of the decomposed sub-questions, revealing that existing MLLMs struggle to produce high-quality sub-questions. To address this limitation, we propose a specific finetuning dataset, DecoVQA+, for enhancing the model’s question decomposition capability. Aiming at enabling models to perform appropriate selective decomposition, we propose an efficient finetuning pipeline. The finetuning pipeline consists of our proposed dataset and a training objective for selective decomposition. Finetuned MLLMs demonstrate significant improvements in the quality of sub-questions and the policy of selective question decomposition. Additionally, the models also achieve higher accuracy with selective decomposition on VQA benchmark datasets.

Traditional benchmarking in NLP typically involves using static held-out test sets. However, this approach often results in an overestimation of performance and lacks the ability to offer comprehensive, interpretable, and dynamic assessments of NLP models. Recently, works like DynaBench (Kiela et al., 2021) and CheckList (Ribeiro et al., 2020) have addressed these limitations through behavioral testing of NLP models with test types generated by a multistep human-annotated pipeline. Unfortunately, manually creating a variety of test types requires much human labor, often at prohibitive cost. In this work, we propose SYNTHEVAL, a hybrid behavioral testing framework that leverages large language models (LLMs) to generate a wide range of test types for a comprehensive evaluation of NLP models. SYNTHEVAL first generates sentences via LLMs using controlled generation, and then identifies challenging examples by comparing the predictions made by LLMs with task-specific NLP models. In the last stage, human experts investigate the challenging examples, manually design templates, and identify the types of failures the taskspecific models consistently exhibit. We apply SYNTHEVAL to two classification tasks, sentiment analysis and toxic language detection, and show that our framework is effective in identifying weaknesses of strong models on these tasks.

Recent advances in Large Language Models (LLMs) have sparked wide interest in validating and comprehending the human-like cognitive-behavioral traits LLMs may capture and convey. These cognitive-behavioral traits include typically Attitudes, Opinions, Values (AOVs). However, measuring AOVs embedded within LLMs remains opaque, and different evaluation methods may yield different results. This has led to a lack of clarity on how different studies are related to each other and how they can be interpreted. This paper aims to bridge this gap by providing a comprehensive overview of recent works on the evaluation of AOVs in LLMs. Moreover, we survey related approaches in different stages of the evaluation pipeline in these works. By doing so, we address the potential and challenges with respect to understanding the model, human-AI alignment, and downstream application in social sciences. Finally, we provide practical insights into evaluation methods, model enhancement, and interdisciplinary collaboration, thereby contributing to the evolving landscape of evaluating AOVs in LLMs.

We present the joint CUNI and LMU submission to the MRL 2024 Shared Task on Multi-lingual Multi-task Information Retrieval. The shared task objective was to explore how we can deploy modern methods in NLP in multi-lingual low-resource settings, tested on two sub-tasks: Named-entity recognition and question answering. Our solutions to the subtasks are based on data acquisition and model adaptation. We compare the performance of our submitted systems with the translate-test approach which proved to be the most useful in the previous edition of the shared task. Our results show that using more data as well as fine-tuning recent multilingual pre-trained models leads to considerable improvements over the translate-test baseline.

Climate change (CC) has attracted increasing attention in NLP in recent years. However, detecting the stance on CC in multimodal data is understudied and remains challenging due to a lack of reliable datasets. To improve the understanding of public opinions and communication strategies, this paper presents MultiClimate, the first open-source manually-annotated stance detection dataset with 100 CC-related YouTube videos and 4,209 frame-transcript pairs. We deploy state-of-the-art vision and language models, as well as multimodal models for MultiClimate stance detection. Results show that text-only BERT significantly outperforms image-only ResNet50 and ViT. Combining both modalities achieves state-of-the-art, 0.747/0.749 in accuracy/F1. Our 100M-sized fusion models also beat CLIP and BLIP, as well as the much larger 9B-sized multimodal IDEFICS and text-only Llama3 and Gemma2, indicating that multimodal stance detection remains challenging for large language models.

Generative artificial intelligence (AI) presents large risks for society when it is used to create fake news. A crucial factor for fake news to go viral on social media is that users share such content. Here, we aim to shed light on the sharing behavior of users across human-generated vs. AI-generated fake news. Specifically, we study: (1) What is the perceived veracity of human-generated fake news vs. AI-generated fake news? (2) What is the user’s willingness to share human-generated fake news vs. AI-generated fake news on social media? (3) What socio-economic characteristics let users fall for AI-generated fake news? To this end, we conducted a pre-registered, online experiment with N= 988 subjects and 20 fake news from the COVID-19 pandemic generated by GPT-4 vs. humans. Our findings show that AI-generated fake news is perceived as less accurate than human-generated fake news, but both tend to be shared equally. Further, several socio-economic factors explain who falls for AI-generated fake news.

The 2022 Russian invasion of Ukraine was accompanied by a large-scale, pro-Russian propaganda campaign on social media. However, the strategy behind the dissemination of propaganda has remained unclear, particularly how the online discourse was strategically shaped by the propagandists’ community. Here, we analyze the strategy of the Twitter community using an inverse reinforcement learning (IRL) approach. Specifically, IRL allows us to model online behavior as a Markov decision process, where the goal is to infer the underlying reward structure that guides propagandists when interacting with users with a supporting or opposing stance toward the invasion. Thereby, we aim to understand empirically whether and how between-user interactions are strategically used to promote the proliferation of Russian propaganda. For this, we leverage a large-scale dataset with 349,455 posts with pro-Russian propaganda from 132,131 users. We show that bots and humans follow a different strategy: bots respond predominantly to pro-invasion messages, suggesting that they seek to drive virality; while messages indicating opposition primarily elicit responses from humans, suggesting that they tend to engage in critical discussions. To the best of our knowledge, this is the first study analyzing the strategy behind propaganda from the 2022 Russian invasion of Ukraine through the lens of IRL.

Online hate speech is responsible for violent attacks such as, e.g., the Pittsburgh synagogue shooting in 2018, thereby posing a significant threat to vulnerable groups and society in general. However, little is known about what makes hate speech on social media go viral. In this paper, we collect N = 25,219 cascades with 65,946 retweets from X (formerly known as Twitter) and classify them as hateful vs. normal. Using a generalized linear regression, we then estimate differences in the spread of hateful vs. normal content based on author and content variables. We thereby identify important determinants that explain differences in the spreading of hateful vs. normal content. For example, hateful content authored by verified users is disproportionally more likely to go viral than hateful content from non-verified ones: hateful content from a verified user (as opposed to normal content) has a 3.5 times larger cascade size, a 3.2 times longer cascade lifetime, and a 1.2 times larger structural virality. Altogether, we offer novel insights into the virality of hate speech on social media.

Registering pre-operative modalities, such as magnetic resonance imaging or computed tomography, to ultrasound images is crucial for guiding clinicians during surgeries and biopsies. Recently, deep-learning approaches have been proposed to increase the speed and accuracy of this registration problem. However, all of these approaches need expensive supervision from the ultrasound domain. In this work, we propose a multitask generative framework that needs weak supervision only from the pre-operative imaging domain during training. To perform a deformable registration, the proposed framework translates a magnetic resonance image to the ultrasound domain while preserving the structural content. To demonstrate the efficacy of the proposed method, we tackle the registration problem of pre-operative 3D MR to transrectal ultrasonography images as necessary for targeted prostate biopsies. We use an in-house dataset of 600 patients, divided into 540 for training, 30 for validation, and the remaining for testing. An expert manually segmented the prostate in both modalities for validation and test sets to assess the performance of our framework. The proposed framework achieves a 3.58 mm target registration error on the expert-selected landmarks, 89.2% in the Dice score, and 1.81 mm 95th percentile Hausdorff distance on the prostate masks in the test set. Our experiments demonstrate that the proposed generative model successfully translates magnetic resonance images into the ultrasound domain. The translated image contains the structural content and fine details due to an ultrasound-specific two-path design of the generative model. The proposed framework enables training learning-based registration methods while only weak supervision from the pre-operative domain is available.

Algorithmic profiling is increasingly used in the public sector with the hope of allocating limited public resources more effectively and objectively. One example is the prediction-based profiling of job seekers to guide the allocation of support measures by public employment services. However, empirical evaluations of potential side-effects such as unintended discrimination and fairness concerns are rare in this context. We systematically compare and evaluate statistical models for predicting job seekers’ risk of becoming long-term unemployed concerning subgroup prediction performance, fairness metrics, and vulnerabilities to data analysis decisions. Focusing on Germany as a use case, we evaluate profiling models under realistic conditions using large-scale administrative data. We show that despite achieving high prediction performance on average, profiling models can be considerably less accurate for vulnerable social subgroups. In this setting, different classification policies can have very different fairness implications. We therefore call for rigorous auditing processes before such models are put to practice.

This study investigates the predictive capability of radiomics in determining programmed cell death ligand 1 (PD-L1) expression (>=1%) status in non-small cell lung cancer (NSCLC) patients using a newly collected [18F]FDG PET/CT dataset. We aimed to replicate and validate the radiomics-based machine learning (ML) model proposed by Zhao et al. [2] predicting PD-L1 status from PET/CT-imaging.
An independent cohort of 254 NSCLC patients underwent [18F]FDG PET/CT imaging, with primary tumor segmentation conducted using lung tissue window (LTW) and more conservative soft tissue window (STW) methods. Radiomics models (“Rad-score” and “complex model”) and a clinical-stage model from Zhao et al. were evaluated via 10-fold cross-validation and AUC analysis, alongside a benchmark-study comparing different ML-model pipelines. Clinicopathological data were collected from medical records.
On our data, the Rad-score model yielded mean AUCs of 0.593 (STW) and 0.573 (LTW), below Zhao et al.’s 0.761. The complex model achieved mean AUCs of 0.505 (STW) and 0.519 (LTW), lower than Zhao et al.’s 0.769. The clinical model showed a mean AUC of 0.555, below Zhao et al.’s 0.64. All models performed significantly lower than Zhao et al.’s findings. Our benchmark study on four ML pipelines revealed consistently low performance across all configurations.
Our study failed to replicate original findings, suggesting poor model performance and questioning predictive value of radiomics features in classifying PD-L1 expression from PET/CT imaging. These results highlight challenges in replicating radiomics-based ML models and stress the need for rigorous validation

Self-supervised learning guided by masked image modeling, such as masked autoencoder (MAE), has attracted wide attention for pretraining vision transformers in remote sensing. However, MAE tends to excessively focus on pixel details, limiting the model’s capacity for semantic understanding, particularly for noisy synthetic aperture radar (SAR) images. In this article, we explore spectral and spatial remote sensing image features as improved MAE-reconstruction targets. We first conduct a study on reconstructing various image features, all performing comparably well or better than raw pixels. Based on such observations, we propose feature guided MAE (FG-MAE): reconstructing a combination of histograms of oriented gradients (HOG) and normalized difference indices (NDI) for multispectral images, and reconstructing HOG for SAR images. Experimental results on three downstream tasks illustrate the effectiveness of FG-MAE with a particular boost for SAR imagery (e.g., up to 5% better than MAE on EuroSAT-SAR). Furthermore, we demonstrate the well-inherited scalability of FG-MAE and release a first series of pretrained vision transformers for medium-resolution SAR and multispectral images.

We present a flexible modelling approach to analyse time-varying exposures and recurrent events in team sports injuries. The approach is based on the piece-wise exponential additive mixed model where the effects of past exposures (i.e. high-intensity training loads) may accumulate over time and present complex forms of association. In order to identify a relevant time window at which past exposures have an impact on the current risk, we propose a penalty approach. We conduct a simulation study to evaluate the performance of the proposed model, under different true weight functions and different levels of heterogeneity between recurrent events. Finally, we illustrate the approach with a case study application involving an elite male football team participating in the Spanish LaLiga competition. The cohort includes time-loss injuries and external training load variables tracked by Global Positioning System devices, during the seasons 2017–2018 and 2018–2019.

Feature attribution methods attempt to explain neural network predictions by identifying relevant features. However, establishing a cohesive framework for assessing feature attribution remains a challenge. There are several views through which we can evaluate attributions. One principal lens is to observe the effect of perturbing attributed features on the model’s behavior (i.e., faithfulness). While providing useful insights, existing faithfulness evaluations suffer from shortcomings that we reveal in this paper. In this work, we propose two new perspectives within the faithfulness paradigm that reveal intuitive properties: soundness and completeness. Soundness assesses the degree to which attributed features are truly predictive features, while completeness examines how well the resulting attribution reveals all the predictive features. The two perspectives are based on a firm mathematical foundation and provide quantitative metrics that are computable through efficient algorithms. We apply these metrics to mainstream attribution methods, offering a novel lens through which to analyze and compare feature attribution methods.

Online hate speech poses a serious threat to individual well-being and societal cohesion. A promising solution to curb online hate speech is counterspeech. Counterspeech is aimed at encouraging users to reconsider hateful posts by direct replies. However, current methods lack scalability due to the need for human intervention or fail to adapt to the specific context of the post. A potential remedy is the use of generative AI, specifically large language models (LLMs), to write tailored counterspeech messages. In this paper, we analyze whether contextualized counterspeech generated by state-of-the-art LLMs is effective in curbing online hate speech. To do so, we conducted a large-scale, pre-registered field experiment (N=2,664) on the social media platform Twitter/X. Our experiment followed a 2x2 between-subjects design and, additionally, a control condition with no counterspeech. On the one hand, users posting hateful content on Twitter/X were randomly assigned to receive either (a) contextualized counterspeech or (b) non-contextualized counterspeech. Here, the former is generated through LLMs, while the latter relies on predefined, generic messages. On the other hand, we tested two counterspeech strategies: (a) promoting empathy and (b) warning about the consequences of online misbehavior. We then measured whether users deleted their initial hateful posts and whether their behavior changed after the counterspeech intervention (e.g., whether users adopted a less toxic language). We find that non-contextualized counterspeech employing a warning-of-consequence strategy significantly reduces online hate speech. However, contextualized counterspeech generated by LLMs proves ineffective and may even backfire.

Multimodal Large Language Models (MLLMs) have significantly advanced visual tasks by integrating visual representations into large language models (LLMs). The textual modality, inherited from LLMs, equips MLLMs with abilities like instruction following and in-context learning. In contrast, the visual modality enhances performance in downstream tasks by leveraging rich semantic content, spatial information, and grounding capabilities. These intrinsic modalities work synergistically across various visual tasks. Our research initially reveals a persistent imbalance between these modalities, with text often dominating output generation during visual instruction tuning. This imbalance occurs when using both full fine-tuning and parameter-efficient fine-tuning (PEFT) methods. We then found that re-balancing these modalities can significantly reduce the number of trainable parameters required, inspiring a direction for further optimizing visual instruction tuning. We introduce Modality Linear Representation-Steering (MoReS) to achieve the goal. MoReS effectively re-balances the intrinsic modalities throughout the model, where the key idea is to steer visual representations through linear transformations in the visual subspace across each model layer. To validate our solution, we composed LLaVA Steering, a suite of models integrated with the proposed MoReS method. Evaluation results show that the composed LLaVA Steering models require, on average, 500 times fewer trainable parameters than LoRA needs while still achieving comparable performance across three visual benchmarks and eight visual question-answering tasks. Last, we present the LLaVA Steering Factory, an in-house developed platform that enables researchers to quickly customize various MLLMs with component-based architecture for seamlessly integrating state-of-the-art models, and evaluate their intrinsic modality imbalance.

Meningeal lymphatic vessels (MLVs) are responsible for the drainage of waste products from the human brain. An impairment in their functionality has been associated with aging as well as brain disorders like multiple sclerosis and Alzheimer’s disease. However, MLVs have only recently been described for the first time in magnetic resonance imaging (MRI), and their ramified structure renders manual segmentation particularly difficult. Further, as there is no consistent notion of their appearance, human-annotated MLV structures contain a high inter-rater variability that most automatic segmentation methods cannot take into account. In this work, we propose a new rater-aware training scheme for the popular nnU-Net model, and we explore rater-based ensembling strategies for accurate and consistent segmentation of MLVs. This enables us to boost nnU-Net’s performance while obtaining explicit predictions in different annotation styles and a rater-based uncertainty estimation. Our final model, MLV2-Net, achieves a Dice similarity coefficient of 0.806 with respect to the human reference standard. The model further matches the human inter-rater reliability and replicates age-related associations with MLV volume.

Finding correspondences between 3D shapes is an important and long-standing problem in computer vision, graphics and beyond. While approaches based on machine learning dominate modern 3D shape matching, almost all existing (learning-based) methods require that at least one of the involved shapes is complete. In contrast, the most challenging and arguably most practically relevant setting of matching partially observed shapes, is currently underexplored. One important factor is that existing datasets contain only a small number of shapes (typically below 100), which are unable to serve data-hungry machine learning approaches, particularly in the unsupervised regime. In addition, the type of partiality present in existing datasets is often artificial and far from realistic. To address these limitations and to encourage research on these relevant settings, we provide a generic and flexible framework for the procedural generation of challenging partial shape matching scenarios. Our framework allows for a virtually infinite generation of partial shape matching instances from a finite set of shapes with complete geometry. Further, we manually create cross-dataset correspondences between seven existing (complete geometry) shape matching datasets, leading to a total of 2543 shapes. Based on this, we propose several challenging partial benchmark settings, for which we evaluate respective state-of-the-art methods as baselines.

Deep neural network ensembles are powerful tools for uncertainty quantification, which have recently been re-interpreted from a Bayesian perspective. However, current methods inadequately leverage second-order information of the loss landscape, despite the recent availability of efficient Hessian approximations. We propose a novel approximate Bayesian inference method that modifies deep ensembles to incorporate Stein Variational Newton updates. Our approach uniquely integrates scalable modern Hessian approximations, achieving faster convergence and more accurate posterior distribution approximations. We validate the effectiveness of our method on diverse regression and classification tasks, demonstrating superior performance with a significantly reduced number of training epochs compared to existing ensemble-based methods, while enhancing uncertainty quantification and robustness against overfitting.

Recent AI advances have enabled multi-modal systems to model and translate diverse information spaces. Extending beyond text and vision, we introduce OneProt, a multi-modal AI for proteins that integrates structural, sequence, alignment, and binding site data. Using the ImageBind framework, OneProt aligns the latent spaces of modality encoders along protein sequences. It demonstrates strong performance in retrieval tasks and surpasses state-of-the-art methods in various downstream tasks, including metal ion binding classification, gene-ontology annotation, and enzyme function prediction. This work expands multi-modal capabilities in protein models, paving the way for applications in drug discovery, biocatalytic reaction planning, and protein engineering.

In this paper, we consider functionals of the form Hα(u)=F(u)+αG(u) with α∈[0,+∞), where u varies in a set U≠∅ (without further structure). We first show that, excluding at most countably many values of α, we have that infH⋆αG=supH⋆αG, where H⋆α:=argminUHα, which is assumed to be non-empty. We further prove a stronger result that concerns the {invariance of the} limiting value of the functional G along minimizing sequences for Hα. This fact in turn implies an unexpected consequence for functionals regularized with uniformly convex norms: excluding again at most countably many values of α, it turns out that for a minimizing sequence, convergence to a minimizer in the weak or strong sense is equivalent.

We consider high-dimensional estimation problems where the number of parameters diverges with the sample size. General conditions are established for consistency, uniqueness, and asymptotic normality in both unpenalized and penalized estimation settings. The conditions are weak and accommodate a broad class of estimation problems, including ones with non-convex and group structured penalties. The wide applicability of the results is illustrated through diverse examples, including generalized linear models, multi-sample inference, and stepwise estimation procedures.

Large language models (LLMs) excel at retrieving information from lengthy text, but their vision-language counterparts (VLMs) face difficulties with hour-long videos, especially for temporal grounding. Specifically, these VLMs are constrained by frame limitations, often losing essential temporal details needed for accurate event localization in extended video content. We propose ReVisionLLM, a recursive vision-language model designed to locate events in hour-long videos. Inspired by human search strategies, our model initially targets broad segments of interest, progressively revising its focus to pinpoint exact temporal boundaries. Our model can seamlessly handle videos of vastly different lengths, from minutes to hours. We also introduce a hierarchical training strategy that starts with short clips to capture distinct events and progressively extends to longer videos. To our knowledge, ReVisionLLM is the first VLM capable of temporal grounding in hour-long videos, outperforming previous state-of-the-art methods across multiple datasets by a significant margin (+2.6% R1@0.1 on MAD).

This study investigates the linguistic understanding of Large Language Models (LLMs) regarding signifier (form) and signified (meaning) by distinguishing two LLM evaluation paradigms: psycholinguistic and neurolinguistic. Traditional psycholinguistic evaluations often reflect statistical biases that may misrepresent LLMs’ true linguistic capabilities. We introduce a neurolinguistic approach, utilizing a novel method that combines minimal pair and diagnostic probing to analyze activation patterns across model layers. This method allows for a detailed examination of how LLMs represent form and meaning, and whether these representations are consistent across languages. Our contributions are three-fold: (1) We compare neurolinguistic and psycholinguistic methods, revealing distinct patterns in LLM assessment; (2) We demonstrate that LLMs exhibit higher competence in form compared to meaning, with the latter largely correlated to the former; (3) We present new conceptual minimal pair datasets for Chinese (COMPS-ZH) and German (COMPS-DE), complementing existing English datasets.

What mechanisms underlie linguistic generalization in large language models (LLMs)? This question has attracted considerable attention, with most studies analyzing the extent to which the language skills of LLMs resemble rules. As of yet, it is not known whether linguistic generalization in LLMs could equally well be explained as the result of analogical processes, which can be formalized as similarity operations on stored exemplars. A key shortcoming of prior research is its focus on linguistic phenomena with a high degree of regularity, for which rule-based and analogical approaches make the same predictions. Here, we instead examine derivational morphology, specifically English adjective nominalization, which displays notable variability. We introduce a new method for investigating linguistic generalization in LLMs: focusing on GPT-J, we fit cognitive models that instantiate rule-based and analogical learning to the LLM training data and compare their predictions on a set of nonce adjectives with those of the LLM, allowing us to draw direct conclusions regarding underlying mechanisms. As expected, rule-based and analogical models explain the predictions of GPT-J equally well for adjectives with regular nominalization patterns. However, for adjectives with variable nominalization patterns, the analogical model provides a much better match. Furthermore, GPT-J’s behavior is sensitive to the individual word frequencies, even for regular forms, a behavior that is consistent with an analogical account of regular forms but not a rule-based one. These findings refute the hypothesis that GPT-J’s linguistic generalization on adjective nominalization involves rules, suggesting similarity operations on stored exemplars as the underlying mechanism. Overall, our study suggests that analogical processes play a bigger role in the linguistic generalization of LLMs than previously thought.

A common challenge in continual learning (CL) is catastrophic forgetting, where the performance on old tasks drops after new, additional tasks are learned. In this paper, we propose a novel framework called ReCL to slow down forgetting in CL. Our framework exploits an implicit bias of gradient-based neural networks due to which these converge to margin maximization points. Such convergence points allow us to reconstruct old data from previous tasks, which we then combine with the current training data. Our framework is flexible and can be applied on top of existing, state-of-the-art CL methods to slow down forgetting. We further demonstrate the performance gain from our framework across a large series of experiments, including different CL scenarios (class incremental, domain incremental, task incremental learning) different datasets (MNIST, CIFAR10), and different network architectures. Across all experiments, we find large performance gains through ReCL. To the best of our knowledge, our framework is the first to address catastrophic forgetting by leveraging models in CL as their own memory buffers.

Increasingly frequent publications in the literature report voice quality differences between depressed patients and controls. Here, we examine the possibility of using voice analysis as an early warning signal for the development of emotion disturbances in young adults. As part of a major interdisciplinary European research project in four countries (ECoWeB), examining the effects of web-based prevention programs to reduce the risk for depression in young adults, we analyzed a large number of acoustic voice characteristics in vocal reports of emotions experienced by the participants on a specific day. We were able to identify a number of significant differences in acoustic cues, particularly with respect to the energy distribution in the voice spectrum, encouraging further research efforts to develop promising non-obtrusive risk indicators in the normal speaking voice. This is particularly important in the case of young adults who are less likely to exhibit standard risk factors for depression such as negative life experiences.

Differential privacy (DP) is a widely used approach for mitigating privacy risks when training machine learning models on sensitive data. DP mechanisms add noise during training to limit the risk of information leakage. The scale of the added noise is critical, as it determines the trade-off between privacy and utility. The standard practice is to select the noise scale to satisfy a given privacy budget ε. This privacy budget is in turn interpreted in terms of operational attack risks, such as accuracy, sensitivity, and specificity of inference attacks aimed to recover information about the training data records. We show that first calibrating the noise scale to a privacy budget ε, and then translating {epsilon} to attack risk leads to overly conservative risk assessments and unnecessarily low utility. Instead, we propose methods to directly calibrate the noise scale to a desired attack risk level, bypassing the step of choosing ε. For a given notion of attack risk, our approach significantly decreases noise scale, leading to increased utility at the same level of privacy. We empirically demonstrate that calibrating noise to attack sensitivity/specificity, rather than ε, when training privacy-preserving ML models substantially improves model accuracy for the same risk level. Our work provides a principled and practical way to improve the utility of privacy-preserving ML without compromising on privacy.

Understanding dynamic 3D scenes is fundamental for various applications, including extended reality (XR) and autonomous driving. Effectively integrating semantic information into 3D reconstruction enables holistic representation that opens opportunities for immersive and interactive applications. We introduce SADG, Segment Any Dynamic Gaussian Without Object Trackers, a novel approach that combines dynamic Gaussian Splatting representation and semantic information without reliance on object IDs. In contrast to existing works, we do not rely on supervision based on object identities to enable consistent segmentation of dynamic 3D objects. To this end, we propose to learn semantically-aware features by leveraging masks generated from the Segment Anything Model (SAM) and utilizing our novel contrastive learning objective based on hard pixel mining. The learned Gaussian features can be effectively clustered without further post-processing. This enables fast computation for further object-level editing, such as object removal, composition, and style transfer by manipulating the Gaussians in the scene. We further extend several dynamic novel-view datasets with segmentation benchmarks to enable testing of learned feature fields from unseen viewpoints. We evaluate SADG on proposed benchmarks and demonstrate the superior performance of our approach in segmenting objects within dynamic scenes along with its effectiveness for further downstream editing tasks.

Automatically and rapidly understanding Earth’s surface is fundamental to our grasp of the living environment and informed decision-making. This underscores the need for a unified system with comprehensive capabilities in analyzing Earth’s surface to address a wide range of human needs. The emergence of multimodal large language models (MLLMs) has great potential in boosting the efficiency and convenience of intelligent Earth observation. These models can engage in human-like conversations, serve as unified platforms for understanding images, follow diverse instructions, and provide insightful feedbacks. In this study, we introduce LHRS-Bot-Nova, an MLLM specialized in understanding remote sensing (RS) images, designed to expertly perform a wide range of RS understanding tasks aligned with human instructions. LHRS-Bot-Nova features an enhanced vision encoder and a novel bridge layer, enabling efficient visual compression and better language-vision alignment. To further enhance RS-oriented vision-language alignment, we propose a large-scale RS image-caption dataset, generated through feature-guided image recaptioning. Additionally, we introduce an instruction dataset specifically designed to improve spatial recognition abilities. Extensive experiments demonstrate superior performance of LHRS-Bot-Nova across various RS image understanding tasks. We also evaluate different MLLM performances in complex RS perception and instruction following using a complicated multi-choice question evaluation benchmark, providing a reliable guide for future model selection and improvement.

Topological correctness plays a critical role in many image segmentation tasks, yet most networks are trained using pixel-wise loss functions, such as Dice, neglecting topological accuracy. Existing topology-aware methods often lack robust topological guarantees, are limited to specific use cases, or impose high computational costs. In this work, we propose a novel, graph-based framework for topologically accurate image segmentation that is both computationally efficient and generally applicable. Our method constructs a component graph that fully encodes the topological information of both the prediction and ground truth, allowing us to efficiently identify topologically critical regions and aggregate a loss based on local neighborhood information. Furthermore, we introduce a strict topological metric capturing the homotopy equivalence between the union and intersection of prediction-label pairs. We formally prove the topological guarantees of our approach and empirically validate its effectiveness on binary and multi-class datasets. Our loss demonstrates state-of-the-art performance with up to fivefold faster loss computation compared to persistent homology methods.

In this paper we propose MA-DV2F: Multi-Agent Dynamic Velocity Vector Field. It is a framework for simultaneously controlling a group of vehicles in challenging environments. DV2F is generated for each vehicle independently and provides a map of reference orientation and speed that a vehicle must attain at any point on the navigation grid such that it safely reaches its target. The field is dynamically updated depending on the speed and proximity of the ego-vehicle to other agents. This dynamic adaptation of the velocity vector field allows prevention of imminent collisions. Experimental results show that MA-DV2F outperforms concurrent methods in terms of safety, computational efficiency and accuracy in reaching the target when scaling to a large number of vehicles.

In the recent past, a popular way of evaluating natural language understanding (NLU), was to consider a model’s ability to perform natural language inference (NLI) tasks. In this paper, we investigate if NLI tasks, that are rarely used for LLM evaluation, can still be informative for evaluating LLMs. Focusing on five different NLI benchmarks across six models of different scales, we investigate if they are able to discriminate models of different size and quality and how their accuracies develop during training. Furthermore, we investigate the extent to which the softmax distributions of models align with human distributions in cases where statements are ambiguous or vague. Overall, our results paint a positive picture for the NLI tasks: we find that they are able to discriminate well between models at various stages of training, yet are not (all) saturated. Furthermore, we find that while the similarity of model distributions with human label distributions increases with scale, it is still much higher than the similarity between two populations of humans, making it a potentially interesting statistic to consider.

Curriculum learning (CL) describes a machine learning training strategy in which samples are gradually introduced into the training process based on their difficulty. Despite a partially contradictory body of evidence in the literature, CL finds popularity in deep learning research due to its promise of leveraging human-inspired curricula to achieve higher model performance. Yet, the subjectivity and biases that follow any necessary definition of difficulty, especially for those found in orderings derived from models or training statistics, have rarely been investigated. To shed more light on the underlying unanswered questions, we conduct an extensive study on the robustness and similarity of the most common scoring functions for sample difficulty estimation, as well as their potential benefits in CL, using the popular benchmark dataset CIFAR-10 and the acoustic scene classification task from the DCASE2020 challenge as representatives of computer vision and computer audition, respectively. We report a strong dependence of scoring functions on the training setting, including randomness, which can partly be mitigated through ensemble scoring. While we do not find a general advantage of CL over uniform sampling, we observe that the ordering in which data is presented for CL-based training plays an important role in model performance. Furthermore, we find that the robustness of scoring functions across random seeds positively correlates with CL performance. Finally, we uncover that models trained with different CL strategies complement each other by boosting predictive power through late fusion, likely due to differences in the learnt concepts. Alongside our findings, we release the aucurriculum toolkit (this https URL), implementing sample difficulty and CL-based training in a modular fashion.

This work introduces the key operating principles for autrainer, our new deep learning training framework for computer audition tasks. autrainer is a PyTorch-based toolkit that allows for rapid, reproducible, and easily extensible training on a variety of different computer audition tasks. Concretely, autrainer offers low-code training and supports a wide range of neural networks as well as preprocessing routines. In this work, we present an overview of its inner workings and key capabilities.

Large-scale multimodal representation learning successfully optimizes for zero-shot transfer at test time. Yet the standard pretraining paradigm (contrastive learning on large amounts of image-text data) does not explicitly encourage representations to support few-shot adaptation. In this work, we propose a simple, but carefully designed extension to multimodal pretraining which enables representations to accommodate additional context. Using this objective, we show that vision-language models can be trained to exhibit significantly increased few-shot adaptation: across 21 downstream tasks, we find up to four-fold improvements in test-time sample efficiency, and average few-shot adaptation gains of over 5%, while retaining zero-shot generalization performance across model scales and training durations. In particular, equipped with simple, training-free, metric-based adaptation mechanisms, our representations easily surpass more complex and expensive optimization-based schemes, vastly simplifying generalization to new domains.

As Artificial Intelligence (AI) continues to advance rapidly, Friendly AI (FAI) has been proposed to advocate for more equitable and fair development of AI. Despite its importance, there is a lack of comprehensive reviews examining FAI from an ethical perspective, as well as limited discussion on its potential applications and future directions. This paper addresses these gaps by providing a thorough review of FAI, focusing on theoretical perspectives both for and against its development, and presenting a formal definition in a clear and accessible format. Key applications are discussed from the perspectives of eXplainable AI (XAI), privacy, fairness and affective computing (AC). Additionally, the paper identifies challenges in current technological advancements and explores future research avenues. The findings emphasise the significance of developing FAI and advocate for its continued advancement to ensure ethical and beneficial AI development.

Employing pre-trained Large Language Models (LLMs) has become the de facto standard in Natural Language Processing (NLP) despite their extensive data requirements. Motivated by the recent surge in research focused on training LLMs with limited data, particularly in low-resource domains and languages, this paper surveys recent transfer learning approaches to optimize model performance in downstream tasks where data is scarce. We first address initial and continued pre-training strategies to better leverage prior knowledge in unseen domains and languages. We then examine how to maximize the utility of limited data during fine-tuning and few-shot learning. The final section takes a task-specific perspective, reviewing models and methods suited for different levels of data scarcity. Our goal is to provide practitioners with practical guidelines for overcoming the challenges posed by constrained data while also highlighting promising directions for future research.

As LLMs are increasingly integrated into user-facing applications, addressing biases that perpetuate societal inequalities is crucial. While much work has gone into measuring or mitigating biases in these models, fewer studies have investigated their origins. Therefore, this study examines the correlation between gender-occupation bias in pre-training data and their manifestation in LLMs, focusing on the Dolma dataset and the OLMo model. Using zero-shot prompting and token co-occurrence analyses, we explore how biases in training data influence model outputs. Our findings reveal that biases present in pre-training data are amplified in model outputs. The study also examines the effects of prompt types, hyperparameters, and instruction-tuning on bias expression, finding instruction-tuning partially alleviating representational bias while still maintaining overall stereotypical gender associations, whereas hyperparameters and prompting variation have a lesser effect on bias expression. Our research traces bias throughout the LLM development pipeline and underscores the importance of mitigating bias at the pretraining stage.

Facade semantic segmentation is a long-standing challenge in photogrammetry and computer vision. Although the last decades have witnessed the influx of facade segmentation methods, there is a lack of comprehensive facade classes and data covering the architectural variability. In ZAHA, we introduce Level of Facade Generalization (LoFG), novel hierarchical facade classes designed based on international urban modeling standards, ensuring compatibility with real-world challenging classes and uniform methods’ comparison. Realizing the LoFG, we present to date the largest semantic 3D facade segmentation dataset, providing 601 million annotated points at five and 15 classes of LoFG2 and LoFG3, respectively. Moreover, we analyze the performance of baseline semantic segmentation methods on our introduced LoFG classes and data, complementing it with a discussion on the unresolved challenges for facade segmentation. We firmly believe that ZAHA shall facilitate further development of 3D facade semantic segmentation methods, enabling robust segmentation indispensable in creating urban digital twins.

We present HI-SLAM2, a geometry-aware Gaussian SLAM system that achieves fast and accurate monocular scene reconstruction using only RGB input. Existing Neural SLAM or 3DGS-based SLAM methods often trade off between rendering quality and geometry accuracy, our research demonstrates that both can be achieved simultaneously with RGB input alone. The key idea of our approach is to enhance the ability for geometry estimation by combining easy-to-obtain monocular priors with learning-based dense SLAM, and then using 3D Gaussian splatting as our core map representation to efficiently model the scene. Upon loop closure, our method ensures on-the-fly global consistency through efficient pose graph bundle adjustment and instant map updates by explicitly deforming the 3D Gaussian units based on anchored keyframe updates. Furthermore, we introduce a grid-based scale alignment strategy to maintain improved scale consistency in prior depths for finer depth details. Through extensive experiments on Replica, ScanNet, and ScanNet++, we demonstrate significant improvements over existing Neural SLAM methods and even surpass RGB-D-based methods in both reconstruction and rendering quality.

AI audits are a key mechanism for responsible AI governance. AI audits have been
proposed in a variety of laws and regulations standardized frameworks and guidelines for
industry best practices as a mechanism to facilitate public trust and accountability for AI system developers and deployers. Though AI auditing for the purpose of compliance and assurance with normative requirements currently lacks defined norms and standardized practices, some systematic assurance AI audit methodologies are emerging that are modelled on financial auditing practices. In the spirit of financial audits which aim to uphold trust in the integrity of the proper function of the financial markets for stakeholders, AI audits, on this line of reasoning, aim to provide assurance to their stakeholders about AI organizations’ ability to govern their algorithms in ways that mitigate harms and uphold human values. Against this backdrop, the nature of the auditing industry is currently evolving. Traditional financial auditing practices are becoming increasingly automated by AI and, given the complexity of some AI-systems themselves and the high degree of assurance that they will require, the future of AI auditing itself will foreseeably be automated. This paper makes a first step toward exploring this picture. I argue that current automated auditing trends run the risk of undermining the justificatory plausibility of auditing as an accountability and trust-facilitating mechanism itself. In particular, I suggest that this leads to a continuous desire for verification, in which the epistemic obscurity of auditing assurance – the nature of the judgment provided auditors – increases and the operational capability of audits to achieve their aims decreases.

Remote sensing projects typically generate large amounts of imagery that can be used to train powerful deep neural networks. However, the amount of labeled images is often small, as remote sensing applications generally require expert labelers. Thus, semi-supervised learning (SSL), i.e., learning with a small pool of labeled and a larger pool of unlabeled data, is particularly useful in this domain. Current SSL approaches generate pseudo-labels from model predictions for unlabeled samples. As the quality of these pseudo-labels is crucial for performance, utilizing additional information to improve pseudo-label quality yields a promising direction. For remote sensing images, geolocation and recording time are generally available and provide a valuable source of information as semantic concepts, such as land cover, are highly dependent on spatiotemporal context, e.g., due to seasonal effects and vegetation zones. In this paper, we propose to exploit spatiotemporal metainformation in SSL to improve the quality of pseudo-labels and, therefore, the final model performance. We show that directly adding the available metadata to the input of the predictor at test time degenerates the prediction quality for metadata outside the spatiotemporal distribution of the training set. Thus, we propose a teacher-student SSL framework where only the teacher network uses metainformation to improve the quality of pseudo-labels on the training set. Correspondingly, our student network benefits from the improved pseudo-labels but does not receive metadata as input, making it invariant to spatiotemporal shifts at test time. Furthermore, we propose methods for encoding and injecting spatiotemporal information into the model and introduce a novel distillation mechanism to enhance the knowledge transfer between teacher and student. Our framework dubbed Spatiotemporal SSL can be easily combined with several state-of-the-art SSL methods, resulting in significant and consistent improvements on the BigEarthNet and EuroSAT benchmarks.

Although large language models(LLMs) show amazing capabilities, among various exciting applications discovered for LLMs fall short in other low-resource languages. Besides, most existing methods depend on large-scale dialogue corpora and thus building systems for dialogue generation in a zero-shot scenario remains a considerable challenge. To address this challenge, we propose a novel end-to-end zero-shot dialogue generation model ChatZero based on cross-lingual code-switching method. First, we construct code-switching language and pseudo-target language with placeholders. Then for cross-lingual semantic transfer, we employ unsupervised contrastive learning to minimize the semantics gap of the source language, code-switching language, and pseudo-target language that are mutually positive examples in the high dimensional semantic space. Experiments on the multilingual DailyDialog and DSTC7-AVSD datasets demonstrate that ChatZero can achieve more than 90% of the original performance under the zero-shot case compared to supervised learning, and achieve state-of-the-art performance compared with other baselines.

Self-supervised learning (SSL) has gained prominence due to the increasing availability of unlabeled data and advances in computational efficiency, leading to revolutionized natural language processing with pre-trained language models like BERT and GPT. Representation learning, a core concept in SSL, aims to reduce data dimensionality while preserving meaningful aspects. Conventional SSL methods typically embed data in Euclidean space. However, recent research has revealed that alternative geometries can hold even richer representations, unlocking more meaningful insights from the data. Motivated by this, we propose two novel methods for integrating Hilbert geometry into self-supervised learning for efficient document embedding. First, we present a method directly incorporating Hilbert geometry into the standard Euclidean contrastive learning framework. Additionally, we propose a multi-view hyperbolic contrastive learning framework contrasting both documents and paragraphs. Our findings demonstrate that contrasting only paragraphs, rather than entire documents, can lead to superior efficiency and effectiveness.

Language-specific evaluation of large language models (LLMs) for multiple-choice question answering (MCQA) is an important means to test their abilities for a multitude of different dimensions. With a data set assembled from questions from the German variant of ‘Who Wants to Be a Millionaire?’ we evaluate a set of German models and ChatGPT concerning factual/commonsense knowledge, syntactic abilities, and logical reasoning, amongst others. We contribute this new MCQA data set, extracted from the show’s episodes and designed to evaluate the ability of models to answer this diverse range of questions. To ensure data quality, we describe our preprocessing, encompassing data cleaning, deduplication, and the creation of stratified splits. Furthermore, we fine-tune a set of German LLMs and prompt ChatGPT to provide baseline results. Our findings reveal that these models achieve (partly) satisfactory performance on questions of lower difficulty levels (≤ 1000 euros). As the difficulty increases, performance steadily declines, highlighting the challenging nature of the later stages of the game. We contribute to the ongoing efforts to advance the capabilities of LLMs in comprehending and answering questions by providing a valuable resource for German MCQA research as well as further insights into the limitations of current LLMs.

The partial label ranking (PLR) problem is a supervised learning scenario where the learner predicts a ranking with ties of the labels for a given input instance. It generalizes the well-known label ranking (LR) problem, which only allows for strict rankings. So far, pre-vious learning approaches for PLR have primarily adapted LR methods to accommodate ties in predictions. This paper proposes using multi-output regression (MOR) to address the PLR problem by treating ranking positions as multivariate targets, an approach that has received little attention in both LR and PLR. To effectively employ this approach, we introduce several post-hoc layers that convert MOR results into a ranking, potentially including ties. This framework produces a range of learning approaches, which we demonstrate in experimental evaluations to be competitive with the current state-of-the-art PLR methods.

Business processes from many domains like manufacturing, healthcare, or business administration suffer from different amounts of uncertainty concerning the execution of individual activities and their order of occurrence. As long as a process is not entirely serial, i.e., there are no forks or decisions to be made along the process execution, we are - in the absence of exhaustive domain knowledge - confronted with the question whether and in what order activities should be executed or left out for a given case and a desired outcome. As the occurrence or non-occurrence of events has substantial implications regarding process key performance indicators like throughput times or scrap rate, there is ample need for assessing and modeling that process-inherent uncertainty. We propose a novel way of handling the uncertainty by leveraging the probabilistic mechanisms of Bayesian Networks to model processes from the structural and temporal information given in event log data and offer a comprehensive evaluation of uncertainty by modelling cases in their entirety. In a thorough analysis of well-established benchmark datasets, we show that our Process-aware Bayesian Network is capable of answering process queries concerned with any unknown process sequence regarding activities and/or attributes enhancing the explainability of processes. Our method can infer execution probabilities of activities at different stages and can query probabilities of certain process outcomes. The key benefit of the Process-aware Query System over existing approaches is the ability to deliver probabilistic, case-diagnostic information about the execution of activities via Bayesian inference.

Process mining solutions aim to improve performance, save resources, and address bottlenecks in organizations. However, success depends on data quality and availability, and existing analyses often lack diverse data for rigorous testing. To overcome this, we propose an interactive web application tool, extending the GEDI Python framework, which creates event datasets that meet specific (meta-)features. It provides diverse benchmark event data by exploring new regions within the feature space, enhancing the range and quality of process mining analyses. This tool improves evaluation quality and helps uncover correlations between meta-features and metrics, ultimately enhancing solution effectiveness.

The abundance of new approaches in process mining and the diversity of processes in the real-world, raises the question of this thesis: How can we create benchmarks, which reliably measure the impact of event data features on process mining evaluation? Developing benchmarks, that employ comprehensive intentional ED and also consider connections between ED characteristic features, methods, and metrics, will support process miners to evaluate methods more efficiently and reliably.

Reliable process information, especially regarding trace durations, is crucial for smooth execution. Without it, maintaining a process becomes costly. While many predictive systems aim to identify inefficiencies, they often focus on individual process instances, missing the global perspective. It is essential not only to detect where delays occur but also to pinpoint specific activity transitions causing them. To address this, we propose CC-HIT (Creating Counterfactuals from High-Impact Transitions), which identifies temporal dependencies across the entire process. By focusing on activity transitions, we provide deeper insights into relational impacts, enabling faster resolution of inefficiencies. CC-HIT highlights the most influential transitions on process performance, offering actionable insights for optimization. We validate this method using the BPIC 2020 dataset, demonstrating its effectiveness compared to existing approaches.

Existing techniques for monocular 3D detection have a serious restriction. They tend to perform well only on a limited set of benchmarks, faring well either on ego-centric car views or on traffic camera views, but rarely on both. To encourage progress, this work advocates for an extended evaluation of 3D detection frameworks across different camera perspectives. We make two key contributions. First, we introduce the CARLA Drone dataset, CDrone. Simulating drone views, it substantially expands the diversity of camera perspectives in existing benchmarks. Despite its synthetic nature, CDrone represents a real-world challenge. To show this, we confirm that previous techniques struggle to perform well both on CDrone and a real-world 3D drone dataset. Second, we develop an effective data augmentation pipeline called GroundMix. Its distinguishing element is the use of the ground for creating 3D-consistent augmentation of a training image. GroundMix significantly boosts the detection accuracy of a lightweight one-stage detector. In our expanded evaluation, we achieve the average precision on par with or substantially higher than the previous state of the art across all tested datasets.

3D scene stylization extends the work of neural style transfer to 3D. A vital challenge in this problem is to maintain the uniformity of the stylized appearance across multiple views. A vast majority of the previous works achieve this by training a 3D model for every stylized image and a set of multi-view images. In contrast, we propose a novel architecture trained on a collection of style images that, at test time, produces real time high-quality stylized novel views. We choose the underlying 3D scene representation for our model as 3D Gaussian splatting. We take the 3D Gaussians and process them using a multi-resolution hash grid and a tiny MLP to obtain stylized views. The MLP is conditioned on different style codes for generalization to different styles during test time. The explicit nature of 3D Gaussians gives us inherent advantages over NeRF-based methods, including geometric consistency and a fast training and rendering regime. This enables our method to be useful for various practical use cases, such as augmented or virtual reality. We demonstrate that our method achieves state-of-the-art performance with superior visual quality on various indoor and outdoor real-world data.

Large language models (LLMs) have recently shown impressive performance on tasks involving reasoning, leading to a lively debate on whether these models possess reasoning capabilities similar to humans. However, despite these successes, the depth of LLMs’ reasoning abilities remains uncertain. This uncertainty partly stems from the predominant focus on task performance, measured through shallow accuracy metrics, rather than a thorough investigation of the models’ reasoning behavior. This paper seeks to address this gap by providing a comprehensive review of studies that go beyond task accuracy, offering deeper insights into the models’ reasoning processes. Furthermore, we survey prevalent methodologies to evaluate the reasoning behavior of LLMs, emphasizing current trends and efforts towards more nuanced reasoning analyses. Our review suggests that LLMs tend to rely on surface-level patterns and correlations in their training data, rather than on sophisticated reasoning abilities. Additionally, we identify the need for further research that delineates the key differences between human and LLM-based reasoning. Through this survey, we aim to shed light on the complex reasoning processes within LLMs.

Multiple choice questions (MCQs) are commonly used to evaluate the capabilities of large language models (LLMs). One common way to evaluate the model response is to rank the candidate answers based on the log probability of the first token prediction. An alternative way is to examine the text output. Prior work has shown that first token probabilities lack robustness to changes in MCQ phrasing, and that first token probabilities do not match text answers for instruction-tuned models. Therefore, in this paper, we investigate the robustness of text answers. We show that the text answers are more robust to question perturbations than the first token probabilities, when the first token answers mismatch the text answers. The difference in robustness increases as the mismatch rate becomes greater. As the mismatch reaches over 50%, the text answer is more robust to option order changes than the debiased first token probabilities using state-of-the-art debiasing methods such as PriDe. Our findings provide further evidence for the benefits of text answer evaluation over first token probability evaluation.

Ultrasound is widely used in medical diagnostics allowing for accessible and powerful imaging but suffers from resolution limitations due to diffraction and the finite aperture of the imaging system, which restricts diagnostic use. The impulse function of an ultrasound imaging system is called the point spread function (PSF), which is convolved with the spatial distribution of reflectors in the image formation process. Recovering high-resolution reflector distributions by removing image distortions induced by the convolution process improves image clarity and detail. Conventionally, deconvolution techniques attempt to rectify the imaging system’s dependent PSF, working directly on the radio-frequency (RF) data. However, RF data is often not readily accessible. Therefore, we introduce a physics-based deconvolution process using a modeled PSF, working directly on the more commonly available B-mode images. By leveraging Implicit Neural Representations (INRs), we learn a continuous mapping from spatial locations to their respective echogenicity values, effectively compensating for the discretized image space. Our contribution consists of a novel methodology for retrieving a continuous echogenicity map directly from a B-mode image through a differentiable physics-based rendering pipeline for ultrasound resolution enhancement. We qualitatively and quantitatively evaluate our approach on synthetic data, demonstrating improvements over traditional methods in metrics such as PSNR and SSIM. Furthermore, we show qualitative enhancements on an ultrasound phantom and an in-vivo acquisition of a carotid artery.

In radiation therapy (RT), an accurate delineation of the regions of interest (ROI) and organs at risk (OAR) allows for a more targeted irradiation with reduced side effects. The current clinical workflow for combined MR-linear accelerator devices (MR-linacs) requires the acquisition of a planning MR volume (MR-P), in which the ROI and OAR are accurately segmented by the clinical team. These segmentation maps (S-P) are transferred to the MR acquired on the day of the RT fraction (MR-Fx) using registration, followed by time-consuming manual corrections. The goal of this paper is to enable accurate automatic segmentation of MR-Fx using S-P without clinical workflow disruption. We propose a novel UNet-based architecture, CloverNet, that takes as inputs MR-Fx and S-P in two separate encoder branches, whose latent spaces are concatenated in the bottleneck to generate an improved segmentation of MP-Fx. CloverNet improves the absolute Dice Score by 3.73% (relative +4.34%, p<0.001) when compared with conventional 3D UNet. Moreover, we believe this approach is potentially applicable to other longitudinal use cases in which a prior segmentation of the ROI is available.

Generally, the small size of public medical imaging datasets coupled with stringent privacy concerns, hampers the advancement of data-hungry deep learning models in medical imaging. This study addresses these challenges for 3D cardiac MRI images in the short-axis view. We propose Latent Diffusion Models that generate synthetic images conditioned on medical attributes, while ensuring patient privacy through differentially private model training. To our knowledge, this is the first work to apply and quantify differential privacy in 3D medical image generation. We pre-train our models on public data and finetune them with differential privacy on the UK Biobank dataset. Our experiments reveal that pre-training significantly improves model performance, achieving a Fréchet Inception Distance (FID) of 26.77 at ϵ=10, compared to 92.52 for models without pre-training. Additionally, we explore the trade-off between privacy constraints and image quality, investigating how tighter privacy budgets affect output controllability and may lead to degraded performance. Our results demonstrate that proper consideration during training with differential privacy can substantially improve the quality of synthetic cardiac MRI images, but there are still notable challenges in achieving consistent medical realism.

Surgical data science (SDS) is a field that analyzes patient data before, during, and after surgery to improve surgical outcomes and skills. However, surgical data is scarce, heterogeneous, and complex, which limits the applicability of existing machine learning methods. In this work, we introduce the novel task of future video generation in laparoscopic surgery. This task can augment and enrich the existing surgical data and enable various applications, such as simulation, analysis, and robot-aided surgery. Ultimately, it involves not only understanding the current state of the operation but also accurately predicting the dynamic and often unpredictable nature of surgical procedures. Our proposed method, VISAGE (VIdeo Synthesis using Action Graphs for Surgery), leverages the power of action scene graphs to capture the sequential nature of laparoscopic procedures and utilizes diffusion models to synthesize temporally coherent video sequences. VISAGE predicts the future frames given only a single initial frame, and the action graph triplets. By incorporating domain-specific knowledge through the action graph, VISAGE ensures the generated videos adhere to the expected visual and motion patterns observed in real laparoscopic procedures. The results of our experiments demonstrate high-fidelity video generation for laparoscopy procedures, which enables various applications in SDS.

Graph-based holistic scene representations facilitate surgical workflow understanding and have recently demonstrated significant success. However, this task is often hindered by the limited availability of densely annotated surgical scene data. In this work, we introduce an end-to-end framework for the generation and optimization of surgical scene graphs on a downstream task. Our approach leverages the flexibility of graph-based spectral clustering and the generalization capability of foundation models to generate unsupervised scene graphs with learnable properties. We reinforce the initial spatial graph with sparse temporal connections using local matches between consecutive frames to predict temporally consistent clusters across a temporal neighborhood. By jointly optimizing the spatiotemporal relations and node features of the dynamic scene graph with the downstream task of phase segmentation, we address the costly and annotation-burdensome task of semantic scene comprehension and scene graph generation in surgical videos using only weak surgical phase labels. Further, by incorporating effective intermediate scene representation disentanglement steps within the pipeline, our solution outperforms the SOTA on the CATARACTS dataset by 8% accuracy and 10% F1 score in surgical workflow recognition.

Topological accuracy in medical image segmentation is a highly important property for downstream applications such as network analysis and flow modeling in vessels or cell counting. Recently, significant methodological advancements have brought well-founded concepts from algebraic topology to binary segmentation. However, these approaches have been underexplored in multi-class segmentation scenarios, where topological errors are common. We propose a general loss function for topologically faithful multi-class segmentation extending the recent Betti matching concept, which is based on induced matchings of persistence barcodes. We project the N-class segmentation problem to N single-class segmentation tasks, which allows us to use 1-parameter persistent homology, making training of neural networks computationally feasible. We validate our method on a comprehensive set of four medical datasets with highly variant topological characteristics. Our loss formulation significantly enhances topological correctness in cardiac, cell, artery-vein, and Circle of Willis segmentation.

Deep learning (DL) methods typically require large datasets to effectively learn data distributions. However, in the medical field, data is often limited in quantity, and acquiring labeled data can be costly. To mitigate this data scarcity, data augmentation techniques are commonly employed. Among these techniques, generative models play a pivotal role in expanding datasets. However, when it comes to ultrasound (US) imaging, the authenticity of generated data often diminishes due to the oversight of ultrasound physics.
We propose a novel approach to improve the quality of generated US images by introducing a physics-based diffusion model that is specifically designed for this image modality. The proposed model incorporates an US-specific scheduler scheme that mimics the natural behavior of sound wave propagation in ultrasound imaging. Our analysis demonstrates how the proposed method aids in modeling the attenuation dynamics in US imaging. We present both qualitative and quantitative results based on standard generative model metrics, showing that our proposed method results in overall more plausible images.

In this work, we introduce Progressive Growing of Patch Size, a resource-efficient implicit curriculum learning approach for dense prediction tasks. Our curriculum approach is defined by growing the patch size during model training, which gradually increases the task’s difficulty. We integrated our curriculum into the nnU-Net framework and evaluated the methodology on all 10 tasks of the Medical Segmentation Decathlon. With our approach, we are able to substantially reduce runtime, computational costs, and emissions of network training compared to classical constant patch size training. In our experiments, the curriculum approach resulted in improved convergence. We are able to outperform standard nnU-Net training, which is trained with constant patch size, in terms of Dice Score on 7 out of 10 MSD tasks while only spending roughly 50% of the original training runtime. To the best of our knowledge, our Progressive Growing of Patch Size is the first successful employment of a sample-length curriculum in the form of patch size in the field of computer vision.

Positron emission tomography (PET) is a well-established functional imaging technique for diagnosing brain disorders. However, PET’s high costs and radiation exposure limit its widespread use. In contrast, magnetic resonance imaging (MRI) does not have these limitations. Although it also captures neurodegenerative changes, MRI is a less sensitive diagnostic tool than PET. To close this gap, we aim to generate synthetic PET from MRI. Herewith, we introduce PASTA, a novel pathology-aware image translation framework based on conditional diffusion models. Compared to the state-of-the-art methods, PASTA excels in preserving both structural and pathological details in the target modality, which is achieved through its highly interactive dual-arm architecture and multi-modal condition integration. A cycle exchange consistency and volumetric generation strategy elevate PASTA’s capability to produce high-quality 3D PET scans. Our qualitative and quantitative results confirm that the synthesized PET scans from PASTA not only reach the best quantitative scores but also preserve the pathology correctly. For Alzheimer’s classification, the performance of synthesized scans improves over MRI by 4%, almost reaching the performance of actual PET.

Physics-inspired regularization is desired for intra-patient image registration since it can effectively capture the biomechanical characteristics of anatomical structures. However, a major challenge lies in the reliance on physical parameters: Parameter estimations vary widely across the literature, and the physical properties themselves are inherently subject-specific. In this work, we introduce a novel data-driven method that leverages hypernetworks to learn the tissue-dependent elasticity parameters of an elastic regularizer. Notably, our approach facilitates the estimation of patient-specific parameters without the need to retrain the network. We evaluate our method on three publicly available 2D and 3D lung CT and cardiac MR datasets. We find that with our proposed subject-specific tissue-dependent regularization, a higher registration quality is achieved across all datasets compared to using a global regularizer.

Ultrasound imaging is challenging to interpret due to non-uniform intensities, low contrast, and inherent artifacts, necessitating extensive training for non-specialists. Advanced representation with clear tissue structure separation could greatly assist clinicians in mapping underlying anatomy and distinguishing between tissue layers. Decomposing an image into semantically meaningful segments is mainly achieved using supervised segmentation algorithms. Unsupervised methods are beneficial, as acquiring large labeled datasets is difficult and costly, but despite their advantages, they still need to be explored in ultrasound. This paper proposes a novel unsupervised deep learning strategy tailored to ultrasound to obtain easily interpretable tissue separations. We integrate key concepts from unsupervised deep spectral methods, which combine spectral graph theory with deep learning methods. We utilize self-supervised transformer features for spectral clustering to generate meaningful segments based on ultrasound-specific metrics and shape and positional priors, ensuring semantic consistency across the dataset. We evaluate our unsupervised deep learning strategy on three ultrasound datasets, showcasing qualitative results across anatomical contexts without label requirements. We also conduct a comparative analysis against other clustering algorithms to demonstrate superior segmentation performance, boundary preservation, and label consistency.

Nuclei semantic segmentation is a key component for advancing machine learning and deep learning applications in digital pathology. However, most existing segmentation models are trained and tested on high-quality data acquired with expensive equipment, such as whole slide scanners, which are not accessible to most pathologists in developing countries. These pathologists rely on low-resource data acquired with low-precision microscopes, smartphones, or digital cameras, which have different characteristics and challenges than high-resource data. Therefore, there is a gap between the state-of-the-art segmentation models and the real-world needs of low-resource settings. This work aims to bridge this gap by presenting the first fully annotated African multi-organ dataset for histopathology nuclei semantic segmentation acquired with a low-precision microscope. We also evaluate state-of-the-art segmentation models, including spectral feature extraction encoder and vision transformer-based models, and stain normalization techniques for color normalization of Hematoxylin and Eosin-stained histopathology slides. Our results provide important insights for future research on nuclei histopathology segmentation with low-resource data.