A | Foundations of Machine Learning

tbd

Stationary distributions of multivariate diffusion processes have recently been proposed as probabilistic models of causal systems in statistics and machine learning. Motivated by these developments, we study stationary multivariate diffusion processes with a sparsely structured drift. Our main result gives a characterization of the conditional independence relations that hold in a stationary distribution. The result draws on a graphical representation of the drift structure and pertains to conditional independence relations that hold generally as a consequence of the drift’s sparsity pattern.

Understanding the structure of neural network loss surfaces, particularly the emergence of low-loss tunnels, is critical for advancing neural network theory and practice. In this paper, we propose a novel approach to directly embed loss tunnels into the loss landscape of neural networks. Exploring the properties of these loss tunnels offers new insights into their length and structure and sheds light on some common misconceptions. We then apply our approach to Bayesian neural networks, where we improve subspace inference by identifying pitfalls and proposing a more natural prior that better guides the sampling procedure.

Additive models (AMs) have sparked a lot of interest in machine learning recently, allowing the incorporation of interpretable structures into a wide range of model classes. Many commonly used approaches to fit a wide variety of potentially complex additive models build on the idea of boosting additive models. While boosted additive models (BAMs) work well in practice, certain theoretical aspects are still poorly understood, including general convergence behavior and what optimization problem is being solved when accounting for the implicit regularizing nature of boosting. In this work, we study the solution paths of BAMs and establish connections with other approaches for certain classes of problems. Along these lines, we derive novel convergence results for BAMs, which yield crucial insights into the inner workings of the method. While our results generally provide reassuring theoretical evidence for the practical use of BAMs, they also uncover some ‘pathologies’ of boosting for certain additive model classes concerning their convergence behavior that require caution in practice. We empirically validate our theoretical findings through several numerical experiments.

The landscape of neural network loss functions is known to be highly complex, and the ability of gradient-based approaches to find well-generalizing solutions to such high-dimensional problems is often considered a miracle. Similarly, Bayesian neural networks (BNNs) inherit this complexity through the model’s likelihood. In applications where BNNs are used to account for weight uncertainty, recent advantages in sampling-based inference (SAI) have shown promising results outperforming other approximate Bayesian inference (ABI) methods. In this work, we analyze the approximate posterior implicitly defined by SAI and uncover key insights into its success. Among other things, we demonstrate how SAI handles symmetries differently than ABI, and examine the role of overparameterization. Further, we investigate the characteristics of approximate posteriors with sampling budgets scaled far beyond previously studied limits and explain why the localized behavior of samplers does not inherently constitute a disadvantage.

Prior-fitted networks (PFNs) have emerged as promising foundation models for prediction from tabular data sets, achieving state-of-the-art performance on small to moderate data sizes without tuning. While PFNs are motivated by Bayesian ideas, they do not provide any uncertainty quantification for predictive means, quantiles, or similar quantities. We propose a principled and efficient method to construct Bayesian posteriors for such estimates based on Martingale Posteriors. Several simulated and real-world data examples are used to showcase the resulting uncertainty quantification of our method in inference applications.

Despite significant recent advances in probabilistic meta-learning, it is common for practitioners to avoid using deep learning models due to a comparative lack of interpretability. Instead, many practitioners simply use non-meta-models such as Gaussian processes with interpretable priors, and conduct the tedious procedure of training their model from scratch for each task they encounter. While this is justifiable for tasks with a limited number of data points, the cubic computational cost of exact Gaussian process inference renders this prohibitive when each task has many observations. To remedy this, we introduce a family of models that meta-learn sparse Gaussian process inference. Not only does this enable rapid prediction on new tasks with sparse Gaussian processes, but since our models have clear interpretations as members of the neural process family, it also allows manual elicitation of priors in a neural process for the first time. In meta-learning regimes for which the number of observed tasks is small or for which expert domain knowledge is available, this offers a crucial advantage.

Large language models (LLMs) have demonstrated remarkable capabilities in natural language processing, yet their effectiveness in handling historical languages remains largely unexplored. This study examines the performance of open-source LLMs in part-of-speech (POS) tagging for Old Occitan, a historical language characterized by non-standardized orthography and significant diachronic variation. Through comparative analysis of two distinct corpora-hagiographical and medical texts-we evaluate how current models handle the inherent challenges of processing a low-resource historical language. Our findings demonstrate critical limitations in LLM performance when confronted with extreme orthographic and syntactic variability. We provide detailed error analysis and specific recommendations for improving model performance in historical language processing. This research advances our understanding of LLM capabilities in challenging linguistic contexts while offering practical insights for both computational linguistics and historical language studies.

There is growing interest in extending average treatment effect (ATE) estimation to incorporate non-tabular data, such as images and text, which may act as sources of confounding. Neglecting these effects risks biased results and flawed scientific conclusions. However, incorporating non-tabular data necessitates sophisticated feature extractors, often in combination with ideas of transfer learning. In this work, we investigate how latent features from pre-trained neural networks can be leveraged to adjust for sources of confounding. We formalize conditions under which these latent features enable valid adjustment and statistical inference in ATE estimation, demonstrating results along the example of double machine learning. In this context, we also discuss critical challenges inherent to latent feature learning and downstream parameter estimation using those. As our results are agnostic to the considered data modality, they represent an important first step towards a theoretical foundation for the usage of latent representation from foundation models in ATE estimation.

Markov Chain Monte Carlo (MCMC) algorithms are widely regarded as the gold standard for approximate inference in Bayesian neural networks (BNNs). However, they remain computationally expensive and prone to inefficiencies, such
as dying samplers, frequently leading to substantial waste of computational resources. While prior work has presented warmstarting techniques as an effective method to mitigate these inefficiencies, we provide a more comprehensive empirical analysis of how initializations of samplers affect their behavior. Based on various experiments examining the dynamics of warmstarting MCMC, we propose novel warmstarting strategies that leverage performance predictors and adaptive termination criteria to achieve better-performing, yet more cost-efficient, models. In numerical experiments, we demonstrate that this approach provides a practical pathway to more resource-efficient approximate inference in BNNs.

We discover a theoretical connection between explanation estimation and distribution compression that significantly improves the approximation of feature attributions, importance, and effects. While the exact computation of various machine learning explanations requires numerous model inferences and becomes impractical, the computational cost of approximation increases with an ever-increasing size of data and model parameters. We show that the standard i.i.d. sampling used in a broad spectrum of algorithms for post-hoc explanation leads to an approximation error worthy of improvement. To this end, we introduce Compress Then Explain (CTE), a new paradigm of sample-efficient explainability. It relies on distribution compression through kernel thinning to obtain a data sample that best approximates its marginal distribution. CTE significantly improves the accuracy and stability of explanation estimation with negligible computational overhead. It often achieves an on-par explanation approximation error 2-3x faster by using fewer samples, i.e. requiring 2-3x fewer model evaluations. CTE is a simple, yet powerful, plug-in for any explanation method that now relies on i.i.d. sampling.

Sparse regularization techniques are well-established in machine learning, yet their application in neural networks remains challenging due to the non-differentiability of penalties like the L1 norm, which is incompatible with stochastic gradient descent. A promising alternative is shallow weight factorization, where weights are decomposed into two factors, allowing for smooth optimization of L1-penalized neural networks by adding differentiable L2 regularization to the factors. In this work, we introduce deep weight factorization, extending previous shallow approaches to more than two factors. We theoretically establish equivalence of our deep factorization with non-convex sparse regularization and analyze its impact on training dynamics and optimization. Due to the limitations posed by standard training practices, we propose a tailored initialization scheme and identify important learning rate requirements necessary for training factorized networks. We demonstrate the effectiveness of our deep weight factorization through experiments on various architectures and datasets, consistently outperforming its shallow counterpart and widely used pruning methods.

Fine-tuned large language models (LLMs) often exhibit overconfidence, particularly when trained on small datasets, resulting in poor calibration and inaccurate uncertainty estimates. Evidential Deep Learning (EDL), an uncertainty-aware approach, enables uncertainty estimation in a single forward pass, making it a promising method for calibrating fine-tuned LLMs. However, despite its computational efficiency, EDL is prone to overfitting, as its training objective can result in overly concentrated probability distributions. To mitigate this, we propose regularizing EDL by incorporating an information bottleneck (IB). Our approach IB-EDL suppresses spurious information in the evidence generated by the model and encourages truly predictive information to influence both the predictions and uncertainty estimates. Extensive experiments across various fine-tuned LLMs and tasks demonstrate that IB-EDL outperforms both existing EDL and non-EDL approaches. By improving the trustworthiness of LLMs, IB-EDL facilitates their broader adoption in domains requiring high levels of confidence calibration.

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.

Despite recent advances, sampling-based inference for Bayesian Neural Networks (BNNs) remains a significant challenge in probabilistic deep learning. While sampling-based approaches do not require a variational distribution assumption, current state-of-the-art samplers still struggle to navigate the complex and highly multimodal posteriors of BNNs. As a consequence, sampling still requires considerably longer inference times than non-Bayesian methods even for small neural networks, despite recent advances in making software implementations more efficient. Besides the difficulty of finding high-probability regions, the time until samplers provide sufficient exploration of these areas remains unpredictable. To tackle these challenges, we introduce an ensembling approach that leverages strategies from optimization and a recently proposed sampler called Microcanonical Langevin Monte Carlo (MCLMC) for efficient, robust and predictable sampling performance. Compared to approaches based on the state-of-the-art No-U-Turn Sampler, our approach delivers substantial speedups up to an order of magnitude, while maintaining or improving predictive performance and uncertainty quantification across diverse tasks and data modalities. The suggested Microcanonical Langevin Ensembles and modifications to MCLMC additionally enhance the method’s predictability in resource requirements, facilitating easier parallelization. All in all, the proposed method offers a promising direction for practical, scalable inference for BNNs.

The correct way to quantify predictive uncertainty in neural networks remains a topic of active discussion. In particular, it is unclear whether the state-of-the art entropy decomposition leads to a meaningful representation of model, or epistemic, uncertainty (EU) in the light of a debate that pits ignorance against disagreement perspectives. We aim to reconcile the conflicting viewpoints by arguing that both are valid but arise from different learning situations. Notably, we show that the presence of shortcuts is decisive for EU manifesting as disagreement.

A core component of modern large language models is the attention mechanism, but its immense parameter count necessitates structured sparsity for resource-efficient optimization and inference. Traditional sparsity penalties, such as the group lasso, are non-smooth and thus incompatible with standard stochastic gradient descent methods. To address this, we propose a deep gating mechanism that reformulates the structured sparsity penalty into a fully differentiable optimization problem, allowing effective and principled norm-based group sparsification without requiring specialized non-smooth optimizers. Our theoretical analysis and empirical results demonstrate that this approach enables structured sparsity with simple stochastic gradient descent or variants while maintaining predictive performance.

The weights of neural networks (NNs) have recently gained prominence as a new data modality in machine learning, with applications ranging from accuracy and hyperparameter prediction to representation learning or weight generation. One approach to leverage NN weights involves training autoencoders (AEs), using contrastive and reconstruction losses. This allows such models to be applied to a wide variety of downstream tasks, and they demonstrate strong predictive performance and low reconstruction error. However, despite the low reconstruction error, these AEs reconstruct NN models with deteriorated performance compared to the original ones, limiting their usability with regard to model weight generation. In this paper, we identify a limitation of weight-space AEs, specifically highlighting that a structural loss, that uses the Euclidean distance between original and reconstructed weights, fails to capture some features critical for reconstructing high-performing models. We analyze the addition of a behavioral loss for training AEs in weight space, where we compare the output of the reconstructed model with that of the original one, given some common input. We show a strong synergy between structural and behavioral signals, leading to increased performance in all downstream tasks evaluated, in particular NN weights reconstruction and generation.

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.

Bias-preserving methods in fairness-aware machine learning (fairML) focus on metrics that prioritize formal equality by balancing error rates across subgroups. These methods can perpetuate historical discrimination embedded in real-world data. In contrast, bias-transforming methods aim for substantive equality by actively addressing historical inequalities. As a contribution to bias-transforming methods, we introduce the concept of privilege scores, a novel approach to identifying and quantifying individual privilege in machine learning tasks. Privilege scores use causal inference techniques to compare real-world outcomes to those in a ‘fair’ world in which the protected attributes do not influence the target variable. This individual-level perspective provides actionable insights for applications such as affirmative action and beyond. Key contributions include (1) the formalization of privilege scores, (2) a methodological framework for estimation with uncertainty quantification via confidence intervals, (3) an interpretable machine learning approach for understanding privilege score contributions, and (4) a novel in-processing method, Multi-PrivScore, to mitigate model-level discrimination during model training. Experiments on simulated and real-world data demonstrate the usefulness of privilege scores. Overall, our work highlights privilege scores as a versatile tool for assessing and mitigating historical discrimination in various machine learning applications.

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.

Many studies suggest that searching for parking is associated with significant direct and indirect costs. Therefore, it is appealing to reduce the time that car drivers spend on finding an available parking spot, especially in urban areas where the space for all road users is limited. The prediction of on-street parking spot occupancy can provide drivers with guidance on where clear parking spaces are likely to be found. This field of research has gained more and more attention in the last decade through the increasing availability of real-time parking spot occupancy data. In this paper, we pursue a statistical approach for the prediction of parking spot occupancy, where we make use of time-to-event models and semi-Markov process theory. The latter involves the employment of Laplace transformations as well as their inversion, which is an ambitious numerical task. We apply our methodology to data from the City of Melbourne in Australia. Our main result is that the semi-Markov model outperforms a Markov model in terms of both true negative rate and true positive rate while this is essentially achieved by respecting the current duration that a parking space already spends in its initial state.

Advances in single-cell ‘-omics’ allow unprecedented insights into the transcriptional profiles of individual cells and, when combined with large-scale perturbation screens, enable measuring of the effect of targeted perturbations on the whole transcriptome. These advances provide an opportunity to better understand the causative role of genes in complex biological processes. In this Perspective, we delineate the application of causal machine learning to single-cell genomics and its associated challenges. We first present the causal model that is most commonly applied to single-cell biology and then identify and discuss potential approaches to three open problems: the lack of generalization of models to novel experimental conditions, the complexity of interpreting learned models, and the difficulty of learning cell dynamics.

Estimating the generalization error (GE) of machine learning models is fundamental, with resampling methods being the most common approach. However, in non-standard settings, particularly those where observations are not independently and identically distributed, resampling using simple random data divisions may lead to biased GE estimates. This paper strives to present well-grounded guidelines for GE estimation in various such non-standard settings: clustered data, spatial data, unequal sampling probabilities, concept drift, and hierarchically structured outcomes. Our overview combines well-established methodologies with other existing methods that, to our knowledge, have not been frequently considered in these particular settings. A unifying principle among these techniques is that the test data used in each iteration of the resampling procedure should reflect the new observations to which the model will be applied, while the training data should be representative of the entire data set used to obtain the final model. Beyond providing an overview, we address literature gaps by conducting simulation studies. These studies assess the necessity of using GE-estimation methods tailored to the respective setting. Our findings corroborate the concern that standard resampling methods often yield biased GE estimates in non-standard settings, underscoring the importance of tailored GE estimation.

Single-cell data provide high-dimensional measurements of the transcriptional states of cells, but extracting insights into the regulatory functions of genes, particularly identifying transcriptional mechanisms affected by biological perturbations, remains a challenge. Many perturbations induce compensatory cellular responses, making it difficult to distinguish direct from indirect effects on gene regulation. Modeling how gene regulatory functions shape the temporal dynamics of these responses is key to improving our understanding of biological perturbations. Dynamical models based on differential equations offer a principled way to capture transcriptional dynamics, but their application to single-cell data has been hindered by computational constraints, stochasticity, sparsity, and noise. Existing methods either rely on low-dimensional representations or make strong simplifying assumptions, limiting their ability to model transcriptional dynamics at scale. We introduce a Functional and Learnable model of Cell dynamicS, FLeCS, that incorporates gene network structure into coupled differential equations to model gene regulatory functions. Given (pseudo)time-series single-cell data, FLeCS accurately infers cell dynamics at scale, provides improved functional insights into transcriptional mechanisms perturbed by gene knockouts, both in myeloid differentiation and K562 Perturb-seq experiments, and simulates single-cell trajectories of A549 cells following small-molecule perturbations.

Empirical substantive research, such as in the life or social sciences, is commonly categorized into the two modes exploratory and confirmatory, both of which are essential to scientific progress. The former is also referred to as hypothesis-generating or data-contingent research, the latter is also called hypothesis-testing research. In the context of empirical methodological research in statistics, however, the exploratory-confirmatory distinction has received very little attention so far. Our paper aims to fill this gap. First, we revisit the concept of empirical methodological research through the lens of the exploratory-confirmatory distinction. Secondly, we examine current practice with respect to this distinction through a literature survey including 115 articles from the field of biostatistics. Thirdly, we provide practical recommendations towards more appropriate design, interpretation, and reporting of empirical methodological research in light of this distinction. In particular, we argue that both modes of research are crucial to methodological progress, but that most published studies – even if sometimes disguised as confirmatory – are essentially of exploratory nature. We emphasize that it may be adequate to consider empirical methodological research as a continuum between ‘pure’ exploration and ‘strict’ confirmation, recommend transparently reporting the mode of conducted research within the spectrum between exploratory and confirmatory, and stress the importance of study protocols written before conducting the study, especially in confirmatory methodological research.

The term ‘researcher degrees of freedom’ (RDF), which was introduced in metascientific literature in the context of the replication crisis in science, refers to the extent of flexibility a scientist has in making decisions related to data analysis. These choices occur at all stages of the data analysis process. In combination with selective reporting, RDF may lead to over-optimistic statements and an increased rate of false positive findings. Even though the concept has been mainly discussed in fields such as epidemiology or psychology, similar problems affect methodological statistical research. Researchers who develop and evaluate statistical methods are left with a multitude of decisions when designing their comparison studies. This leaves room for an over-optimistic representation of the performance of their preferred method(s). The present paper defines and explores a particular RDF that has not been previously identified and discussed. When interpreting the results of real data examples that are most often part of methodological evaluations, authors typically tell a domain-specific ‘story’ that best supports their argumentation in favor of their preferred method. However, there are often plenty of other plausible stories that would support different conclusions. We define the ‘storytelling fallacy’ as the selective use of anecdotal domain-specific knowledge to support the superiority of specific methods in real data examples. While such examples fed by domain knowledge play a vital role in methodological research, if deployed inappropriately they can also harm the validity of conclusions on the investigated methods. The goal of our work is to create awareness for this issue, fuel discussions on the role of real data in generating evidence in methodological research and warn readers of methodological literature against naive interpretations of real data examples.

Cell tracking is a key computational task in live-cell microscopy, but fully automated analysis of high-throughput imaging requires reliable and, thus, uncertainty-aware data analysis tools, as the amount of data recorded within a single experiment exceeds what humans are able to overlook. We here propose and benchmark various methods to reason about and quantify uncertainty in linear assignment-based cell tracking algorithms. Our methods take inspiration from statistics and machine learning, leveraging two perspectives on the cell tracking problem explored throughout this work: Considering it as a Bayesian inference problem and as a classification problem. Our methods admit a framework-like character in that they equip any frame-to-frame tracking method with uncertainty quantification. We demonstrate this by applying it to various existing tracking algorithms including the recently presented Transformer-based trackers. We demonstrate empirically that our methods yield useful and well-calibrated tracking uncertainties.

Exposure assessment in occupational epidemiology may involve multiple unknown quantities that are measured or reconstructed simultaneously for groups of workers and over several years. Additionally, exposures may be collected using different assessment strategies, depending on the period of exposure. As a consequence, researchers who are analyzing occupational cohort studies are commonly faced with challenging structures of exposure measurement error, involving complex dependence structures and multiple measurement error models, depending on the period of exposure. However, previous work has often made many simplifying assumptions concerning these errors. In this work, we propose a Bayesian hierarchical approach to account for a broad range of error structures arising in occupational epidemiology. The considered error structures may involve several unknown quantities that can be subject to mixtures of Berkson and classical measurement error. It is possible to account for different error structures, depending on the exposure period and the location of a worker. Moreover, errors can present complex dependence structures over time and between workers. We illustrate the proposed hierarchical approach on a subgroup of the German cohort of uranium miners to account for potential exposure uncertainties in the association between radon exposure and lung cancer mortality. The performance of the proposed approach and its sensitivity to model misspecification are evaluated in a simulation study. The results show that biases in estimates arising from very complex measurement errors can be corrected through the proposed Bayesian hierarchical approach.

Variable selection is an important step in the analysis of high-dimensional data, yet there are limited options for survival outcomes in the presence of competing risks. Commonly employed penalized Cox regression considers each event type separately through cause-specific models, neglecting possibly shared information between them. We adapt the feature-weighted elastic net (fwelnet), an elastic net generalization, to survival outcomes and competing risks. For two causes, our proposed algorithm fits two alternating cause-specific models, where each model receives the coefficient vector of the complementary model as prior information. We dub this ‘‘cooperative penalized regression’’, as it enables the modeling of competing risk data with cause-specific models while accounting for shared effects between causes. Coefficients that are shrunken toward zero in the model for the first cause will receive larger penalization weights in the model for the second cause and vice versa. Through multiple iterations, this process ensures stronger penalization of uninformative predictors in both models. We demonstrate our method’s variable selection capabilities on simulated genomics data and apply it to bladder cancer microarray data. We evaluate selection performance using the positive predictive value for the correct selection of informative features and the false positive rate for the selection of uninformative variables. The benchmark compares results with cause-specific penalized Cox regression, random survival forests, and likelihood-boosted Cox regression. Results indicate that our approach is more effective at selecting informative features and removing uninformative features. In settings without shared effects, variable selection performance is similar to cause-specific penalized Cox regression.

Gene set analysis, a popular approach for analyzing high-throughput gene expression data, aims to identify sets of genes that show enriched expression patterns between two conditions. In addition to the multitude of methods available for this task, users are typically left with many options when creating the required input and specifying the internal parameters of the chosen method. This flexibility can lead to uncertainty about the “right” choice, further reinforced by a lack of evidence-based guidance. Especially when their statistical experience is scarce, this uncertainty might entice users to produce preferable results using a ’trial-and-error’ approach. While it may seem unproblematic at first glance, this practice can be viewed as a form of ‘cherry-picking’ and cause an optimistic bias, rendering the results nonreplicable on independent data. After this problem has attracted a lot of attention in the context of classical hypothesis testing, we now aim to raise awareness of such overoptimism in the different and more complex context of gene set analyses. We mimic a hypothetical researcher who systematically selects the analysis variants yielding their preferred results, thereby considering three distinct goals they might pursue. Using a selection of popular gene set analysis methods, we tweak the results in this way for two frequently used benchmark gene expression data sets. Our study indicates that the potential for overoptimism is particularly high for a group of methods frequently used despite being commonly criticized. We conclude by providing practical recommendations to counter overoptimism in research findings in gene set analysis and beyond.

Recent machine-learning (ML)-based advances in single-cell data science have enabled the stratification of human tissue donors at single-cell resolution, promising to provide valuable diagnostic and prognostic insights. However, such insights are susceptible to biases. Here we discuss various biases that emerge along the pipeline of ML-based single-cell analysis, ranging from societal biases affecting whose samples are collected, to clinical and cohort biases that influence the generalizability of single-cell datasets, biases stemming from single-cell sequencing, ML biases specific to (weakly supervised or unsupervised) ML models trained on human single-cell samples and biases during the interpretation of results from ML models. We end by providing methods for single-cell data scientists to assess and mitigate biases, and call for efforts to address the root causes of biases.

Many two-sample problems call for a comparison of two distributions from an exponential family. Density ratio estimation methods provide ways to solve such problems through direct estimation of the differences in natural parameters. The term direct indicates that one avoids estimating both marginal distributions. In this context, we consider the Kullback–Leibler Importance Estimation Procedure (KLIEP), which has been the subject of recent work on differential networks. Our main result shows that the existence of the KLIEP estimator is characterized by whether the average sufficient statistic for one sample belongs to the convex hull of the set of all sufficient statistics for data points in the second sample. For high-dimensional problems it is customary to regularize the KLIEP loss by adding the product of a tuning parameter and a norm of the vector of parameter differences. We show that the existence of the regularized KLIEP estimator requires the tuning parameter to be no less than the dual norm-based distance between the average sufficient statistic and the convex hull. The implications of these existence issues are explored in applications to differential network analysis.

Positivity violations pose a key challenge in the estimation of causal effects, particularly for continuous interventions. Current approaches for addressing this issue include the use of projection functions or modified treatment policies. While effective in many contexts, these methods can result in estimands that potentially do not align well with the original research question, thereby leading to compromises in interpretability. In this paper, we introduce a novel diagnostic tool, the non-overlap ratio, to detect positivity violations. To address these violations while maintaining interpretability, we propose a data-adaptive solution, specially a ‘most feasible’ intervention strategy. Our strategy operates on a unit-specific basis. For a given intervention of interest, we first assess whether the intervention value is feasible for each unit. For units with sufficient support, conditional on confounders, we adhere to the intervention of interest. However, for units lacking sufficient support, as identified through the assessment of the non-overlap ratio, we do not assign the actual intervention value of interest. Instead, we assign the closest feasible value within the support region. We propose an estimator using g-computation coupled with flexible conditional density estimation to estimate high- and low support regions to estimate this new estimand. Through simulations, we demonstrate that our method effectively reduces bias across various scenarios by addressing positivity violations. Moreover, when positivity violations are absent, the method successfully recovers the standard estimand. We further validate its practical utility using real-world data from the CHAPAS-3 trial, which enrolled HIV-positive children in Zambia and Uganda.

Bias-transforming methods of fairness-aware machine learning aim to correct a non-neutral status quo with respect to a protected attribute (PA). Current methods, however, lack an explicit formulation of what drives non-neutrality. We introduce privilege scores (PS) to measure PA-related privilege by comparing the model predictions in the real world with those in a fair world in which the influence of the PA is removed. At the individual level, PS can identify individuals who qualify for affirmative action; at the global level, PS can inform bias-transforming policies. After presenting estimation methods for PS, we propose privilege score contributions (PSCs), an interpretation method that attributes the origin of privilege to mediating features and direct effects. We provide confidence intervals for both PS and PSCs. Experiments on simulated and real-world data demonstrate the broad applicability of our methods and provide novel insights into gender and racial privilege in mortgage and college admissions applications.

Large ensembles of climate models are indispensable for analyzing natural climate variability and estimating the occurrence of rare extreme events. Many hydrometeorological applications—such as compound event analysis, return period estimation, weather forecasting, downscaling, and bias correction—rely on an accurate representation of the multivariate distribution of climate variables. However, at high temporal resolutions, variables like precipitation often exhibit significant zero-inflation and heavy-tailed distributions. This inflation propagates through the entire multivariate dependence structure, complicating the relationships between zero-inflated and non-inflated variables. Inadequate modeling and correction of these dependencies can substantially degrade the reliability of hydrometeorological methodologes.
In an earlier work, we developed a novel multivariate density decomposition for zero inflated variables based on vine copulas. This method has been integrated into multivariate Vine Copula Bias Correction for partially zero-inflated margins (VBC), with potential applications in other fields facing high-resolution climate data challenges. We resume the idea behind VBC and illustrate it’s advantages to other bias correction methods. This highlights the interpretability and the advantages of control and assessment of the results generated by VBC.

Empirical human-AI alignment aims to make AI systems act in line with observed human behavior. While noble in its goals, we argue that empirical alignment can inadvertently introduce statistical biases that warrant caution. This position paper thus advocates against naive empirical alignment, offering prescriptive alignment and a posteriori empirical alignment as alternatives. We substantiate our principled argument by tangible examples like human-centric decoding of language models.

Factor analysis models explain dependence among observed variables by a smaller number of unobserved factors. A main challenge in confirmatory factor analysis is determining whether the factor loading matrix is identifiable from the observed covariance matrix. The factor loading matrix captures the linear effects of the factors and, if unrestricted, can only be identified up to an orthogonal transformation of the factors. However, in many applications the factor loadings exhibit an interesting sparsity pattern that may lead to identifiability up to column signs. We study this phenomenon by connecting sparse factor models to bipartite graphs and providing sufficient graphical conditions for identifiability of the factor loading matrix up to column signs. In contrast to previous work, our main contribution, the matching criterion, exploits sparsity by operating locally on the graph structure, thereby improving existing conditions. Our criterion is efficiently decidable in time that is polynomial in the size of the graph, when restricting the search steps to sets of bounded size.

Out-of-distribution generalization of machine learning models remains challenging since the models are inherently bound to the training data distribution. This especially manifests, when the learned models rely on spurious correlations. Most of the existing approaches apply data manipulation, representation learning, or learning strategies to achieve generalizable models. Unfortunately, these approaches usually require multiple training domains, group labels, specialized augmentation, or pre-processing to reach generalizable models. We propose a novel approach that addresses these limitations by providing a technique to guide the neural network through the training phase. We first establish input pairs, representing the spurious attribute and describing the invariance, a characteristic that should not affect the outcome of the model. Based on these pairs, we form a corrective gradient complementing the traditional gradient descent approach. We further make this correction mechanism adaptive based on a predefined invariance condition. Experiments on ColoredMNIST, Waterbird-100, and CelebA datasets demonstrate the effectiveness of our approach and the robustness to group shifts.

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.

Quantitative bias analysis (QBA) permits assessment of the expected impact of various imperfections of the available data on the results and conclusions of a particular real-world study. This article extends QBA methodology to multivariable time-to-event analyses with right-censored endpoints, possibly including time-varying exposures or covariates. The proposed approach employs data-driven simulations, which preserve important features of the data at hand while offering flexibility in controlling the parameters and assumptions that may affect the results. First, the steps required to perform data-driven simulations are described, and then two examples of real-world time-to-event analyses illustrate their implementation and the insights they may offer. The first example focuses on the omission of an important time-invariant predictor of the outcome in a prognostic study of cancer mortality, and permits separating the expected impact of confounding bias from noncollapsibility. The second example assesses how imprecise timing of an interval-censored event—ascertained only at sparse times of clinic visits—affects its estimated association with a time-varying drug exposure. The simulation results also provide a basis for comparing the performance of two alternative strategies for imputing the unknown event times in this setting. The R scripts that permit the reproduction of our examples are provided.

We address the computational barrier of deploying advanced deep learning segmentation models in clinical settings by studying the efficacy of network compression through tensor decomposition. We propose a post-training Tucker factorization that enables the decomposition of pre-existing models to reduce computational requirements without impeding segmentation accuracy. We applied Tucker decomposition to the convolutional kernels of the TotalSegmentator (TS) model, an nnU-Net model trained on a comprehensive dataset for automatic segmentation of 117 anatomical structures. Our approach reduced the floating-point operations (FLOPs) and memory required during inference, offering an adjustable trade-off between computational efficiency and segmentation quality. This study utilized the publicly available TS dataset, employing various downsampling factors to explore the relationship between model size, inference speed, and segmentation performance. The application of Tucker decomposition to the TS model substantially reduced the model parameters and FLOPs across various compression rates, with limited loss in segmentation accuracy. We removed up to 88% of the model’s parameters with no significant performance changes in the majority of classes after fine-tuning. Practical benefits varied across different graphics processing unit (GPU) architectures, with more distinct speed-ups on less powerful hardware. Post-hoc network compression via Tucker decomposition presents a viable strategy for reducing the computational demand of medical image segmentation models without substantially sacrificing accuracy. This approach enables the broader adoption of advanced deep learning technologies in clinical practice, offering a way to navigate the constraints of hardware capabilities.

Training machine learning models for fair decisions faces two key challenges: The fairness-accuracy trade-off results from enforcing fairness which weakens its predictive performance in contrast to an unconstrained model. The incompatibility of different fairness metrics poses another trade-off – also known as the impossibility theorem. Recent work identifies the bias within the observed data as a possible root cause and shows that fairness and predictive performance are in fact in accord when predictive performance is measured on unbiased data. We offer a causal explanation for these findings using the framework of the FiND (fictitious and normatively desired) world, a ‘fair’ world, where protected attributes have no causal effects on the target variable. We show theoretically that (i) classical fairness metrics deemed to be incompatible are naturally satisfied in the FiND world, while (ii) fairness aligns with high predictive performance. We extend our analysis by suggesting how one can benefit from these theoretical insights in practice, using causal pre-processing methods that approximate the FiND world. Additionally, we propose a method for evaluating the approximation of the FiND world via pre-processing in practical use cases where we do not have access to the FiND world. In simulations and empirical studies, we demonstrate that these pre-processing methods are successful in approximating the FiND world and resolve both trade-offs. Our results provide actionable solutions for practitioners to achieve fairness and high predictive performance simultaneously.

Transformers have emerged as the dominant architecture in the field of deep learning, with a broad range of applications and remarkable in-context learning (ICL) capabilities. While not yet fully understood, ICL has already proved to be an intriguing phenomenon, allowing transformers to learn in context – without requiring further training. In this paper, we further advance the understanding of ICL by demonstrating that transformers can perform full Bayesian inference for commonly used statistical models in context. More specifically, we introduce a general framework that builds on ideas from prior fitted networks and continuous normalizing flows which enables us to infer complex posterior distributions for methods such as generalized linear models and latent factor models. Extensive experiments on real-world datasets demonstrate that our ICL approach yields posterior samples that are similar in quality to state-of-the-art MCMC or variational inference methods not operating in context.

Proposed in Hyvärinen (2005), score matching is a parameter estimation procedure that does not require computation of distributional normalizing constants. In this work we utilize the geometric median of means to develop a robust score matching procedure that yields consistent parameter estimates in settings where the observed data has been contaminated. A special appeal of the proposed method is that it retains convexity in exponential family models. The new method is therefore particularly attractive for non-Gaussian, exponential family graphical models where evaluation of normalizing constants is intractable. Support recovery guarantees for such models when contamination is present are provided. Additionally, support recovery is studied in numerical experiments and on a precipitation dataset. We demonstrate that the proposed robust score matching estimator performs comparably to the standard score matching estimator when no contamination is present but greatly outperforms this estimator in a setting with contamination.

Generalization of machine learning models can be severely compromised by data poisoning, where adversarial changes are applied to the training data. This vulnerability has led to interest in certifying (i.e., proving) that such changes up to a certain magnitude do not affect test predictions. We, for the first time, certify Graph Neural Networks (GNNs) against poisoning attacks, including backdoors, targeting the node features of a given graph. Our certificates are white-box and based upon (i) the neural tangent kernel, which characterizes the training dynamics of sufficiently wide networks; and (ii) a novel reformulation of the bilevel optimization describing poisoning as a mixed-integer linear program. We note that our framework is more general and constitutes the first approach to derive white-box poisoning certificates for NNs, which can be of independent interest beyond graph-related tasks.

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.

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.

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.

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.

The phenomenon of different deep learning models producing similar data representations has garnered significant attention, raising the question of why such representational similarity occurs. Identifiability theory offers a partial explanation: for a broad class of discriminative models, including many popular in representation learning, those assigning equal likelihood to the observations yield representations that are equal up to a linear transformation, if a suitable diversity condition holds. In this work, we identify two key challenges in applying identifiability theory to explain representational similarity. First, the assumption of exact likelihood equality is rarely satisfied by practical models trained with different initializations. To address this, we describe how the representations of two models deviate from being linear transformations of each other, based on their difference in log-likelihoods. Second, we demonstrate that even models with similar and near-optimal loss values can produce highly dissimilar representations due to an underappreciated difference between loss and likelihood. Our findings highlight key open questions and point to future research directions for advancing the theoretical understanding of representational similarity.

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 detection procedure 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.

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.

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.

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.

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.

Metal oxide sensor-based electronic nose (E-Nose) technology provides an easy to use method for breath analysis by detection of volatile organic compound (VOC)-induced changes of electrical conductivity. Resulting signal patterns are then analyzed by machine learning (ML) algorithms. This study aimed to establish breath analysis by E-Nose technology as a diagnostic tool for severe acute respiratory syndrome coronavirus type 2 (SARS-CoV-2) pneumonia within a multi-analyst experiment. Breath samples of 126 subjects with (n = 63) or without SARS-CoV-2 pneumonia (n = 63) were collected using the ReCIVA® Breath Sampler, enriched and stored on Tenax sorption tubes, and analyzed using an E-Nose unit with 10 sensors. ML approaches were applied by three independent data analyst teams and included a wide range of classifiers, hyperparameters, training modes, and subsets of training data. Within the multi-analyst experiment, all teams successfully classified individuals as infected or uninfected with an averaged area under the curve (AUC) larger than 90% and misclassification error lower than 19%, and identified the same sensor as most relevant to classification success. This new method using VOC enrichment and E-Nose analysis combined with ML can yield results similar to polymerase chain reaction (PCR) detection and superior to point-of-care (POC) antigen testing. Reducing the sensor set to the most relevant sensor may prove interesting for developing targeted POC testing.

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.

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.

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.

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.

Functional data analysis (FDA) is a statistical framework that allows for the analysis of curves, images, or functions on higher dimensional domains. The goals of FDA, such as descriptive analyses, classification, and regression, are generally the same as for statistical analyses of scalar-valued or multivariate data, but FDA brings additional challenges due to the high- and infinite dimensionality of observations and parameters, respectively. This paper provides an introduction to FDA, including a description of the most common statistical analysis techniques, their respective software implementations, and some recent developments in the field. The paper covers fundamental concepts such as descriptives and outliers, smoothing, amplitude and phase variation, and functional principal component analysis. It also discusses functional regression, statistical inference with functional data, functional classification and clustering, and machine learning approaches for functional data analysis. The methods discussed in this paper are widely applicable in fields such as medicine, biophysics, neuroscience, and chemistry, and are increasingly relevant due to the widespread use of technologies that allow for the collection of functional data. Sparse functional data methods are also relevant for longitudinal data analysis. All presented methods are demonstrated using available software in R by analyzing a data set on human motion and motor control. To facilitate the understanding of the methods, their implementation, and hands-on application, the code for these practical examples is made available on Github.

Climate model large ensembles are an essential research tool for analysing and quantifying natural climate variability and providing robust information for rare extreme events. The models simulated representations of reality are susceptible to bias due to incomplete understanding of physical processes. This paper aims to correct the bias of five climate variables from the CRCM5 Large Ensemble over Central Europe at a 3-hourly temporal resolution. At this high temporal resolution, two variables, precipitation and radiation, exhibit a high share of zero inflation. We propose a novel bias-correction method, VBC (Vine copula bias correction), that models and transfers multivariate dependence structures for zero-inflated margins in the data from its error-prone model domain to a reference domain. VBC estimates the model and reference distribution using vine copulas and corrects the model distribution via (inverse) Rosenblatt transformation. To deal with the variables’ zero-inflated nature, we develop a new vine density decomposition that accommodates such variables and employs an adequately randomized version of the Rosenblatt transform. This novel approach allows for more accurate modelling of multivariate zero-inflated climate data. Compared with state-of-the-art correction methods, VBC is generally the best-performing correction and the most accurate method for correcting zero-inflated events.

Open-ended text generation has become a prominent task in natural language processing due to the rise of powerful (large) language models. However, evaluating the quality of these models and the employed decoding strategies remains challenging because of trade-offs among widely used metrics such as coherence, diversity, and perplexity. Decoding methods often excel in some metrics while underperforming in others, complicating the establishment of a clear ranking. In this paper, we present novel ranking strategies within this multicriteria framework. Specifically, we employ benchmarking approaches based on partial orderings and present a new summary metric designed to balance existing automatic indicators, providing a more holistic evaluation of text generation quality. Furthermore, we discuss the alignment of these approaches with human judgments. Our experiments demonstrate that the proposed methods offer a robust way to compare decoding strategies, exhibit similarities with human preferences, and serve as valuable tools in guiding model selection for open-ended text generation tasks. Finally, we suggest future directions for improving evaluation methodologies in text generation. Our codebase, datasets, and models are publicly available.

Causal discovery amounts to learning a directed acyclic graph (DAG) that encodes a causal model. This model selection problem can be challenging due to its large combinatorial search space, particularly when dealing with non-parametric causal models. Recent research has sought to bypass the combinatorial search by reformulating causal discovery as a continuous optimization problem, employing constraints that ensure the acyclicity of the graph. In non-parametric settings, existing approaches typically rely on finite-dimensional approximations of the relationships between nodes, resulting in a score-based continuous optimization problem with a smooth acyclicity constraint. In this work, we develop an alternative approximation method by utilizing reproducing kernel Hilbert spaces (RKHS) and applying general sparsity-inducing regularization terms based on partial derivatives. Within this framework, we introduce an extended RKHS representer theorem. To enforce acyclicity, we advocate the log-determinant formulation of the acyclicity constraint and show its stability. Finally, we assess the performance of our proposed RKHS-DAGMA procedure through simulations and illustrative data analyses.

A fundamental challenge of scientific research is inferring causal relations based on observed data. One commonly used approach involves utilizing structural causal models that postulate noisy functional relations among interacting variables. A directed graph naturally represents these models and reflects the underlying causal structure. However, classical identifiability results suggest that, without conducting additional experiments, this causal graph can only be identified up to a Markov equivalence class of indistinguishable models. Recent research has shown that focusing on linear relations with equal error variances can enable the identification of the causal structure from mere observational data. Nonetheless, practitioners are often primarily interested in the effects of specific interventions, rendering the complete identification of the causal structure unnecessary. In this work, we investigate the extent to which less restrictive assumptions of partial homoscedasticity are sufficient for identifying the causal effects of interest. Furthermore, we construct mathematically rigorous confidence regions for total causal effects under structure uncertainty and explore the performance gain of relying on stricter error assumptions in a simulation study.

We study the robustness of global post-hoc explanations for predictive models trained on tabular data. Effects of predictor features in black-box supervised learning are an essential diagnostic tool for model debugging and scientific discovery in applied sciences. However, how vulnerable they are to data and model perturbations remains an open research question. We introduce several theoretical bounds for evaluating the robustness of partial dependence plots and accumulated local effects. Our experimental results with synthetic and real-world datasets quantify the gap between the best and worst-case scenarios of (mis)interpreting machine learning predictions globally.

Self-contrastive learning has proven effective for vision and natural language tasks. It aims to learn aligned data representations by encoding similar and dissimilar sentence pairs without human annotation. Therefore, data augmentation plays a crucial role in the learned embedding quality. However, in natural language processing (NLP), creating augmented samples for unsupervised contrastive learning is challenging since random editing may modify the semantic meanings of sentences and thus affect learning good representations. In this paper, we introduce a simple, still effective approach dubbed ADD (Attention-Driven Dropout) to generate better-augmented views of sentences to be used in self-contrastive learning. Given a sentence and a Pre-trained Transformer Language Model (PLM), such as RoBERTa, we use the aggregated attention scores of the PLM to remove the less “informative” tokens from the input. We consider two alternative algorithms based on NAIVEAGGREGATION across layers/heads and ATTENTIONROLLOUT [1]. Our approach significantly improves the overall performance of various self-supervised contrastive-based methods, including SIMCSE [14], DIFFCSE [10], and INFOCSE [33] by facilitating the generation of high-quality positive pairs required by these methods. Through empirical evaluations on multiple Semantic Textual Similarity (STS) and Transfer Learning tasks, we observe enhanced performance across the board.

Ensembling a neural network is a widely recognized approach to enhance model performance, estimate uncertainty, and improve robustness in deep supervised learning. However, deep ensembles often come with high computational costs and memory demands. In addition, the efficiency of a deep ensemble is related to diversity among the ensemble members, which is challenging for large, over-parameterized deep neural networks. Moreover, ensemble learning has not yet seen such widespread adoption for unsupervised learning and it remains a challenging endeavor for self-supervised or unsupervised representation learning. Motivated by these challenges, we present a novel self-supervised training regime that leverages an ensemble of independent sub-networks, complemented by a new loss function designed to encourage diversity. Our method efficiently builds a sub-model ensemble with high diversity, leading to well-calibrated estimates of model uncertainty, all achieved with minimal computational overhead compared to traditional deep self-supervised ensembles. To evaluate the effectiveness of our approach, we conducted extensive experiments across various tasks, including in-distribution generalization, out-of-distribution detection, dataset corruption, and semi-supervised settings. The results demonstrate that our method significantly improves prediction reliability. Our approach not only achieves excellent accuracy but also enhances calibration, improving on important baseline performance across a wide range of self-supervised architectures in computer vision, natural language processing, and genomics data.

Background: Random forests have become popular for clinical risk prediction modeling. In a case study on predicting ovarian malignancy, we observed training AUCs close to 1. Although this suggests overfitting, performance was competitive on test data. We aimed to understand the behavior of random forests for probability estimation by (1) visualizing data space in three real-world case studies and (2) a simulation study.
Methods: For the case studies, multinomial risk estimates were visualized using heatmaps in a 2-dimensional subspace. The simulation study included 48 logistic data-generating mechanisms (DGM), varying the predictor distribution, the number of predictors, the correlation between predictors, the true AUC, and the strength of true predictors. For each DGM, 1000 training datasets of size 200 or 4000 with binary outcomes were simulated, and random forest models were trained with minimum node size 2 or 20 using the ranger R package, resulting in 192 scenarios in total. Model performance was evaluated on large test datasets (N = 100,000).
Results: The visualizations suggested that the model learned “spikes of probability” around events in the training set. A cluster of events created a bigger peak or plateau (signal), isolated events local peaks (noise). In the simulation study, median training AUCs were between 0.97 and 1 unless there were 4 binary predictors or 16 binary predictors with a minimum node size of 20. The median discrimination loss, i.e., the difference between the median test AUC and the true AUC, was 0.025 (range 0.00 to 0.13). Median training AUCs had Spearman correlations of around 0.70 with discrimination loss. Median test AUCs were higher with higher events per variable, higher minimum node size, and binary predictors. Median training calibration slopes were always above 1 and were not correlated with median test slopes across scenarios (Spearman correlation − 0.11). Median test slopes were higher with higher true AUC, higher minimum node size, and higher sample size.
Conclusions: Random forests learn local probability peaks that often yield near perfect training AUCs without strongly affecting AUCs on test data. When the aim is probability estimation, the simulation results go against the common recommendation to use fully grown trees in random forest models.

Background: Predictive modeling based on multi-omics data, which incorporates several types of omics data for the same patients, has shown potential to outperform single-omics predictive modeling. Most research in this domain focuses on incorporating numerous data types, despite the complexity and cost of acquiring them. The prevailing assumption is that increasing the number of data types necessarily improves predictive performance. However, the integration of less informative or redundant data types could potentially hinder this performance. Therefore, identifying the most effective combinations of omics data types that enhance predictive performance is critical for cost-effective and accurate predictions.
Methods: In this study, we systematically evaluated the predictive performance of all 31 possible combinations including at least one of five genomic data types (mRNA, miRNA, methylation, DNAseq, and copy number variation) using 14 cancer datasets with right-censored survival outcomes, publicly available from the TCGA database. We employed various prediction methods and up-weighted clinical data in every model to leverage their predictive importance. Harrell’s C-index and the integrated Brier Score were used as performance measures. To assess the robustness of our findings, we performed a bootstrap analysis at the level of the included datasets. Statistical testing was conducted for key results, limiting the number of tests to ensure a low risk of false positives.
Results: Contrary to expectations, we found that using only mRNA data or a combination of mRNA and miRNA data was sufficient for most cancer types. For some cancer types, the additional inclusion of methylation data led to improved prediction results. Far from enhancing performance, the introduction of more data types most often resulted in a decline in performance, which varied between the two performance measures.
Conclusions: Our findings challenge the prevailing notion that combining multiple omics data types in multi-omics survival prediction improves predictive performance. Thus, the widespread approach in multi-omics prediction of incorporating as many data types as possible should be reconsidered to avoid suboptimal prediction results and unnecessary expenditure.

Objective. This study aimed to develop convolutional neural networks (CNNs) models to predict the energy expenditure (EE) of children from raw accelerometer data. Additionally, this study sought to external validation of the CNN models in addition to the linear regression (LM), random forest (RF), and full connected neural network (FcNN) models published in Steenbock et al (2019 J. Meas. Phys. Behav. 2 94–102). Approach. Included in this study were 41 German children (3.0–6.99 years) for the training and internal validation who were equipped with GENEActiv, GT3X+, and activPAL accelerometers. The external validation dataset consisted of 39 Canadian children (3.0–5.99 years) that were equipped with OPAL, GT9X, GENEActiv, and GT3X+ accelerometers. EE was recorded simultaneously in both datasets using a portable metabolic unit. The protocols consisted of a semi-structured activities ranging from low to high intensities. The root mean square error (RMSE) values were calculated and used to evaluate model performances. Main results. (1) The CNNs outperformed the LM (13.17%–23.81% lower mean RMSE values), FcNN (8.13%–27.27% lower RMSE values) and the RF models (3.59%–18.84% lower RMSE values) in the internal dataset. (2) In contrast, it was found that when applied to the external Canadian dataset, the CNN models had consistently higher RMSE values compared to the LM, FcNN, and RF. Significance. Although CNNs can enhance EE prediction accuracy, their ability to generalize to new datasets and accelerometer brands/models, is more limited compared to LM, RF, and FcNN models.

In prediction tasks with multi-class outcomes, identifying covariates specifically associated with one or more outcome classes can be important. Conventional variable importance measures (VIMs) from random forests (RFs), like permutation and Gini importance, focus on overall predictive performance or node purity, without differentiating between the classes. Therefore, they can be expected to fail to distinguish class-associated covariates from covariates that only distinguish between groups of classes. We introduce a VIM called multi-class VIM, tailored for identifying exclusively class-associated covariates, via a novel RF variant called multi forests (MuFs). The trees in MuFs use both multi-way and binary splitting. The multi-way splits generate child nodes for each class, using a split criterion that evaluates how well these nodes represent their respective classes. This setup forms the basis of the multi-class VIM, which measures the discriminatory ability of the splits performed in the respective covariates with regard to this split criterion. Alongside the multi-class VIM, we introduce a second VIM, the discriminatory VIM. This measure, based on the binary splits, assesses the strength of the general influence of the covariates, irrespective of their class-associatedness. Simulation studies demonstrate that the multi-class VIM specifically ranks class-associated covariates highly, unlike conventional VIMs which also rank other types of covariates highly. Analyses of 121 datasets reveal that MuFs often have slightly lower predictive performance compared to conventional RFs. This is, however, not a limiting factor given the algorithm’s primary purpose of calculating the multi-class VIM.

When assessing the quality of prediction models in machine learning, confidence intervals (CIs) for the generalization error, which measures predictive performance, are a crucial tool. Luckily, there exist many methods for computing such CIs and new promising approaches are continuously being proposed. Typically, these methods combine various resampling procedures, most popular among them cross-validation and bootstrapping, with different variance estimation techniques. Unfortunately, however, there is currently no consensus on when any of these combinations may be most reliably employed and how they generally compare. In this work, we conduct the first large-scale study comparing CIs for the generalization error - empirically evaluating 13 different methods on a total of 18 tabular regression and classification problems, using four different inducers and a total of eight loss functions. We give an overview of the methodological foundations and inherent challenges of constructing CIs for the generalization error and provide a concise review of all 13 methods in a unified framework. Finally, the CI methods are evaluated in terms of their relative coverage frequency, width, and runtime. Based on these findings, we are able to identify a subset of methods that we would recommend. We also publish the datasets as a benchmarking suite on OpenML and our code on GitHub to serve as a basis for further studies.

To reduce the need for human annotations, large language models (LLMs) have been proposed as judges of the quality of other candidate models. LLM judges are typically evaluated by measuring the correlation with human judgments on generation tasks such as summarization or machine translation. In contrast, we study LLM judges on mathematical reasoning tasks. These tasks require multi-step reasoning, and the correctness of their solutions is verifiable, enabling a more objective evaluation. We perform a detailed performance analysis and find that the used judges are mostly unable to improve task performance but are able to pick the better model. Our analysis uncovers a strong correlation between judgment performance and the candidate model task performance. We observe that judges tend to choose the model of higher quality even if its answer is incorrect. Further, we show that it is possible to use statistics, such as the task performances of the individual models, to predict judgment performance. In an ablation, we either swap or mask the candidate answers and observe that judges often keep the original judgment, providing evidence that judges incorporate writing style in their judgments. In summary, we find that regularities in the judgments are quantifiable using statistical measures and provide various angles on exploiting them.

With the usage of tremendous amounts of text data for training powerful large language models such as ChatGPT, the issue of analysing and securing data quality has become more pressing than ever. Any biases, stereotypes and discriminatory patterns that exist in the training data can be reproduced, reinforced or broadly disseminated by the models in production. Therefore, it is crucial to carefully select and monitor the text data that is used as input to train the model. Due to the vast amount of training data, this process needs to be (at least partially) automated. In this work, we introduce a novel approach for automatically detecting gender discrimination in text data on the actor level based on linguistic discourse analysis. Specifically, we combine existing information extraction (IE) techniques to partly automate the qualitative research done in linguistic discourse analysis. We focus on two important steps: Identifying the respectiveperson-named-entity (an actor) and all forms it is referred to (Nomination), and detecting the characteristics it is ascribed (Predication). Asa proof of concept, we integrate these two steps into a pipeline for automated text analysis. The separate building blocks of the pipeline could be flexibly adapted, extended, and scaled for bigger datasets to accommodate a wide range of usage scenarios and specific ML tasks or help social scientists with analysis tasks. We showcase and evaluate our approach on several real and simulated exemplary texts.

Automatic correction of errors in Handwritten Text Recognition (HTR) output poses persistent challenges yet to be fully resolved. In this study, we introduce a shared task aimed at addressing this challenge, which attracted 271 submissions, yielding only a handful of promising approaches. This paper presents the datasets, the most effective methods, and an experimental analysis in error-correcting HTRed manuscripts and papyri in Byzantine Greek, the language that followed Classical and preceded Modern Greek. By using recognised and transcribed data from seven centuries, the two best-performing methods are compared, one based on a neural encoder-decoder architecture and the other based on engineered linguistic rules. We show that the recognition error rate can be reduced by both, up to 2.5 points at the level of characters and up to 15 at the level of words, while also elucidating their respective strengths and weaknesses.

In this work, we present a collaboratively and continuously developed open-source educational resource (OSER) for teaching natural language processing at two different universities. We shed light on the principles we followed for the initial design of the course and the rationale for ongoing developments, followed by a reflection on the inter-university collaboration for designing and maintaining teaching material. When reflecting on the latter, we explicitly emphasize the considerations that need to be made when facing heterogeneous groups and when having to accommodate multiple examination regulations within one single course framework. Relying on the fundamental principles of OSER developments as defined by Bothmann et al. (2023) proved to be an important guideline during this process. The final part pertains to open-sourcing our teaching material, coping with the increasing speed of developments in the field, and integrating the course digitally, also addressing conflicting priorities and challenges we are currently facing.

Bayesian inference in deep neural networks is challenging due to the high-dimensional, strongly multi-modal parameter posterior density landscape. Markov chain Monte Carlo approaches asymptotically recover the true posterior but are considered prohibitively expensive for large modern architectures. Local methods, which have emerged as a popular alternative, focus on specific parameter regions that can be approximated by functions with tractable integrals. While these often yield satisfactory empirical results, they fail, by definition, to account for the multi-modality of the parameter posterior. In this work, we argue that the dilemma between exact-but-unaffordable and cheap-but-inexact approaches can be mitigated by exploiting symmetries in the posterior landscape. Such symmetries, induced by neuron interchangeability and certain activation functions, manifest in different parameter values leading to the same functional output value. We show theoretically that the posterior predictive density in Bayesian neural networks can be restricted to a symmetry-free parameter reference set. By further deriving an upper bound on the number of Monte Carlo chains required to capture the functional diversity, we propose a straightforward approach for feasible Bayesian inference. Our experiments suggest that efficient sampling is indeed possible, opening up a promising path to accurate uncertainty quantification in deep learning.

With the increased use of machine learning (ML) models within automated decision-making systems, the demands on the quality of ML models are growing. Pure prediction quality is no longer the sole quality criterion; in particular, there is an increasing demand to consider fairness aspects. This paper pursues two goals. First, it summarizes the current fairness discussion in the field of ML (fairML) and describes the most recent developments, especially with respect to the philosophical foundations of the concept of fairness within ML. On the other hand, the question is addressed to what extent so-called ‘extra-legal’ characteristics may be used in the compilation of qualified rent indices. A recent proposal by Kauermann and Windmann (AStA Wirtschafts- und Sozialstatistisches Archiv, Volume 17, 2023) on using extra-legal features in qualified rent indices includes a model-based imputation method, which we contrast with the legal requirements. Finally, we show which alternatives from the field of fairML could be used and outline the different basic philosophical assumptions behind the various methods.

Distributed statistical analyses provide a promising approach for privacy protection when analyzing data distributed over several databases. Instead of directly operating on data, the analyst receives anonymous summary statistics, which are combined into an aggregated result. Further, in discrimination model (prognosis, diagnosis, etc.) development, it is key to evaluate a trained model w.r.t. to its prognostic or predictive performance on new independent data. For binary classification, quantifying discrimination uses the receiver operating characteristics (ROC) and its area under the curve (AUC) as aggregation measure. We are interested to calculate both as well as basic indicators of calibration-in-the-large for a binary classification task using a distributed and privacy-preserving approach…

Cancer cells and pathogens can evade T cell receptors (TCRs) via mutations in immunogenic epitopes. TCR cross-reactivity (i.e., recognition of multiple epitopes with sequence similarities) can counteract such escape but may cause severe side effects in cell-based immunotherapies through targeting self-antigens. To predict the effect of epitope point mutations on T cell functionality, we here present the random forest-based model Predicting T Cell Epitope-Specific Activation against Mutant Versions (P-TEAM). P-TEAM was trained and tested on three datasets with TCR responses to single-amino-acid mutations of the model epitope SIINFEKL, the tumor neo-epitope VPSVWRSSL, and the human cytomegalovirus antigen NLVPMVATV, totaling 9,690 unique TCR-epitope interactions. P-TEAM was able to accurately classify T cell reactivities and quantitatively predict T cell functionalities for unobserved single-point mutations and unseen TCRs. Overall, P-TEAM provides an effective computational tool to study T cell responses against mutated epitopes.

Hintergrund: Die medizinische Codierung von radiologischen Befunden ist essenziell für eine gute Qualität der Versorgung und die korrekte Abrechnung, gleichzeitig aber eine aufwändige und fehleranfällige Aufgabe.
Ziel der Arbeit: Bewertung der Anwendbarkeit natürlicher Sprachverarbeitung (Natural Language Processing, NLP) für die ICD-10-Codierung von radiologischen Befunden in deutscher Sprache durch Finetuning geeigneter Sprachmodelle.
Material und Methoden: In dieser retrospektiven Studie wurden alle Magnetresonanztomographie(MRT)-Befunde unseres Instituts zwischen 2010 und 2020 berücksichtigt. Die ICD-10-Codes bei Entlassung wurden den jeweiligen Befunden zugeordnet, um einen Datensatz für eine Multiclass-Klassifizierung zu erstellen. Finetuning von GermanBERT und flanT5 wurde auf dem Gesamtdatensatz (dstotal) mit 1035 verschiedenen ICD-10-Codes und zwei reduzierten Datensätzen mit den 100 (ds100) und 50 (ds50) häufigsten Codes durchgeführt. Die Performance der Modelle wurde mit Top-k-Genauigkeit für k = 1, 3, 5 evaluiert. In einer Ablationsstudie wurden beide Modelle einmal auf den zugehörigen Metadaten und dem Befund allein trainiert.
Ergebnisse: Der Gesamtdatensatz bestand aus 100.672 radiologischen Befunden, die reduzierten Datensätze ds100 aus 68.103 und ds50 aus 52.293 Berichten. Die Modellperformance stieg, wenn mehrere der besten Voraussagen des Modells in Betracht gezogen wurden, die Anzahl der Zielklassen reduziert wurde und die Metadaten mit dem Befund kombiniert wurden. FlanT5 übertraf GermanBERT in allen Datensätzen und Metriken und eignet sich am besten als medizinischer Codierungsassistent, wobei eine Top-3-Genauigkeit von fast 70% im realitätsnahen Datensatz dstotal erreicht wurde.
Schlussfolgerung: Finetuning von Sprachmodellen verspricht eine zuverlässige Vorhersage von ICD-10-Codes deutscher radiologischer MRT-Befunde in unterschiedlichen Szenarien. Als Codierungsassistent kann flanT5 medizinischen Codierern helfen, informierte Entscheidungen zu treffen und potenziell ihre Arbeitsbelastung reduzieren.

The localization of objects is essential in many applications, such as robotics, virtual and augmented reality, and warehouse logistics. Recent advancements in deep learning have enabled localization using monocular cameras. Traditionally, structure from motion (SfM) techniques predict an object’s absolute position from a point cloud, while absolute pose regression (APR) methods use neural networks to understand the environment semantically. However, both approaches face challenges from environmental factors like motion blur, lighting changes, repetitive patterns, and featureless areas. This study addresses these challenges by incorporating additional information and refining absolute pose estimates with relative pose regression (RPR) methods. RPR also struggles with issues like motion blur. To overcome this, we compute the optical flow between consecutive images using the Lucas–Kanade algorithm and use a small recurrent convolutional network to predict relative poses. Combining absolute and relative poses is difficult due to differences between global and local coordinate systems. Current methods use pose graph optimization (PGO) to align these poses. In this work, we propose recurrent fusion networks to better integrate absolute and relative pose predictions, enhancing the accuracy of absolute pose estimates. We evaluate eight different recurrent units and create a simulation environment to pre-train the APR and RPR networks for improved generalization. Additionally, we record a large dataset of various scenarios in a challenging indoor environment resembling a warehouse with transportation robots. Through hyperparameter searches and experiments, we demonstrate that our recurrent fusion method outperforms PGO in effectiveness.

Machine Learning is a core building block in novel data-driven applications. Practitioners face many ambiguous design decisions while developing practical machine learning (ML) solutions. Automated machine learning (AutoML) facilitates the development of machine learning applications by providing efficient methods for optimizing hyperparameters, searching for neural architectures, or constructing whole ML pipelines (Hutter et al., 2019). Thereby, design decisions such as the choice of modelling, pre-processing, and training algorithm are crucial to obtaining well-performing solutions. By automatically obtaining ML solutions, AutoML aims to lower the barrier to leveraging machine learning and reduce the time needed to develop or adapt ML solutions for new domains or data.
Highly performant software packages for automatically building ML pipelines given data, so-called AutoML systems, are available and can be used off-the-shelf. Typically, AutoML systems evaluate ML models sequentially to return a well-performing single best model or multiple models combined into an ensemble. Existing AutoML systems are typically highly engineered monolithic software developed for specific use cases to perform well and robustly under various conditions…

Human motion analysis and biomechanics are fundamental in a clinical environment, and together, they provide relevant and precise information towards diagnosing numerous neurodegenerative conditions such as stroke, Parkinson’s disease, Alzheimer’s disease, multiple sclerosis, etc. In most neurological disorders, walking is commonly impacted, where performance, quantity, and quality are affected. Thus, motion analysis aims at understanding the cause of altered motion patterns, mainly assisting with the prevention, identification, and rehabilitation. Usually, motion analysis assessment relies on the patient’s self-report and the practitioner’s visually assessed observations. Therefore, such assessments are often subjective and susceptible to human-induced error. In contrast, sophisticated devices can provide quantitative accuracy by equipping practitioners with precise, reliable, and objective measurements to simultaneously monitor an extensive set of parameters for gait analysis (e.g., 3D joint kinematics, muscle activation patterns, muscle forces, and coordination patterns). This book chapter addresses the challenges and describes the technological solutions considered when moving out of the lab condition to the real-world environments, in this case, the clinical setting.

We consider linear non-Gaussian structural equation models that involve latent confounding. In this setting, the causal structure is identifiable, but, in general, it is not possible to identify the specific causal effects. Instead, a finite number of different causal effects result in the same observational distribution. Most existing algorithms for identifying these causal effects use overcomplete independent component analysis (ICA), which often suffers from convergence to local optima. Furthermore, the number of latent variables must be known a priori. To address these issues, we propose an algorithm that operates recursively rather than using overcomplete ICA. The algorithm first infers a source, estimates the effect of the source and its latent parents on their descendants, and then eliminates their influence from the data. For both source identification and effect size estimation, we use rank conditions on matrices formed from higher-order cumulants. We prove asymptotic correctness under the mild assumption that locally, the number of latent variables never exceeds the number of observed variables. Simulation studies demonstrate that our method achieves comparable performance to overcomplete ICA even though it does not know the number of latents in advance.

A growing body of literature in fairness-aware ML aspires to mitigate machine learning (ML)-related unfairness in automated decision-making (ADM) by defining metrics that measure the fairness of an ML model and by proposing methods that ensure that trained ML models achieve low values in those metrics (see, e.g., Verma & Rubin, 2018, Caton & Haas, 2023). However, the underlying concept of fairness, i.e., the question of what fairness is, is rarely discussed, leaving a considerable gap between centuries of philosophical discussion and the recent adoption of the concept in the ML community. We bridge this gap by formalizing a consistent concept of fairness and translating the philosophical considerations into a formal framework for training and evaluating ML models in ADM systems (Bothmann et al., 2024). We argue why and how causal considerations are necessary when assessing fairness in the presence of protected attributes (PAs) by proposing a fictitious, normatively desired (FiND) world where the PAs have no (direct or indirect) causal effect on the target. In practice, this unknown FiND world must be approximated by a warped world, for which the causal effects of the PAs must be removed from the real-world data. We propose rank-preserving interventional distributions to define an estimand of this FiND world and a warping method for estimation (Bothmann et al., 2023). Evaluation criteria for both the method and the resulting ML model are presented. Experiments on simulated data show that our method effectively identifies the most discriminated individuals and mitigates unfairness. Experiments on real-world data showcase the practical application of our method.

This work presents the first large-scale neutral benchmark experiment focused on single-event, right-censored, low-dimensional survival data. Benchmark experiments are essential in methodological research to scientifically compare new and existing model classes through proper empirical evaluation. Existing benchmarks in the survival literature are often narrow in scope, focusing, for example, on high-dimensional data. Additionally, they may lack appropriate tuning or evaluation procedures, or are qualitative reviews, rather than quantitative comparisons. This comprehensive study aims to fill the gap by neutrally evaluating a broad range of methods and providing generalizable conclusions. We benchmark 18 models, ranging from classical statistical approaches to many common machine learning methods, on 32 publicly available datasets. The benchmark tunes for both a discrimination measure and a proper scoring rule to assess performance in different settings. Evaluating on 8 survival metrics, we assess discrimination, calibration, and overall predictive performance of the tested models. Using discrimination measures, we find that no method significantly outperforms the Cox model. However, (tuned) Accelerated Failure Time models were able to achieve significantly better results with respect to overall predictive performance as measured by the right-censored log-likelihood. Machine learning methods that performed comparably well include Oblique Random Survival Forests under discrimination, and Cox-based likelihood-boosting under overall predictive performance. We conclude that for predictive purposes in the standard survival analysis setting of low-dimensional, right-censored data, the Cox Proportional Hazards model remains a simple and robust method, sufficient for practitioners.

When it comes to computation, it is often said that high-dimensional data is particularly challenging, known as the curse of dimensionality. For example, in their seminal work, Beyer et al [1] study the impact of high-dimensional data on nearest neighbor computation. They show that in a wide range of settings, including IID data, the difference between the distance to the nearest neighbor and the distance to the most distant neighbor vanishes as the dimension increases. However, it is arguably often overlooked that they also point out that this result does not hold in certain situations, in particular when the intrinsic dimension of the data is low and/or when the data is distributed in well separable subsets. More generally, it is probably less well known that high dimensionality can make computation easier, to the extent that Kainen [2] even speaks of a blessing of dimensionality. Given these different aspects, a natural question to ask is: when is high dimensionality a curse and when is it not (or even a blessing)? In this talk we approach this question from a geometric point of view. Focusing on the aspect of nearest neighbor (and hence distance) computation, we show that high-dimensional data need not be more challenging than low-dimensional data in many practically relevant situations. In particular, using results from extensive experiments on synthetic and real data, we show that this can be the case for both outlier detection and cluster analysis, and for a range of different data types, including image and functional data [3, 4]. Moreover, based on concepts from manifold learning and topological data analysis, we show that these observations can be explained using a common conceptual foundation.

Neural additive models (NAMs) enhance the transparency of deep neural networks by handling input features in separate additive sub-networks. However, they lack inherent mechanisms that provide calibrated uncertainties and enable selection of relevant features and interactions. Approaching NAMs from a Bayesian perspective, we augment them in three primary ways, namely by a) providing credible intervals for the individual additive sub-networks; b) estimating the marginal likelihood to perform an implicit selection of features via an empirical Bayes procedure; and c) facilitating the ranking of feature pairs as candidates for second-order interaction in fine-tuned models. In particular, we develop Laplace-approximated NAMs (LA-NAMs), which show improved empirical performance on tabular datasets and challenging real-world medical tasks.

We warn against a common but incomplete understanding of empirical research in machine learning (ML) that leads to non-replicable results, makes findings unreliable, and threatens to undermine progress in the field. To overcome this alarming situation, we call for more awareness of the plurality of ways of gaining knowledge experimentally but also of some epistemic limitations. In particular, we argue most current empirical ML research is fashioned as confirmatory research while it should rather be considered exploratory.

Publications proposing novel machine learning methods are often primarily rated by exhibited predictive performance on selected problems. In this position paper we argue that predictive performance alone is not a good indicator for the worth of a publication. Using it as such even fosters problems like inefficiencies of the machine learning research community as a whole and setting wrong incentives for researchers. We therefore put out a call for the publication of “negative” results, which can help alleviate some of these problems and improve the scientific output of the machine learning research community. To substantiate our position, we present the advantages of publishing negative results and provide concrete measures for the community to move towards a paradigm where their publication is normalized.

Automated machine learning (AutoML) was formed around the fundamental objectives of automatically and efficiently configuring machine learning (ML) workflows, aiding the research of new ML algorithms, and contributing to the democratization of ML by making it accessible to a broader audience. Over the past decade, commendable achievements in AutoML have primarily focused on optimizing predictive performance. This focused progress, while substantial, raises questions about how well AutoML has met its broader, original goals. In this position paper, we argue that a key to unlocking AutoML’s full potential lies in addressing the currently underexplored aspect of user interaction with AutoML systems, including their diverse roles, expectations, and expertise. We envision a more human-centered approach in future AutoML research, promoting the collaborative design of ML systems that tightly integrates the complementary strengths of human expertise and AutoML methodologies.

In the current landscape of deep learning research, there is a predominant emphasis on achieving high predictive accuracy in supervised tasks involving large image and language datasets. However, a broader perspective reveals a multitude of overlooked metrics, tasks, and data types, such as uncertainty, active and continual learning, and scientific data, that demand attention. Bayesian deep learning (BDL) constitutes a promising avenue, offering advantages across these diverse settings. This paper posits that BDL can elevate the capabilities of deep learning. It revisits the strengths of BDL, acknowledges existing challenges, and highlights some exciting research avenues aimed at addressing these obstacles. Looking ahead, the discussion focuses on possible ways to combine large-scale foundation models with BDL to unlock their full potential.

The complexity of black-box algorithms can lead to various challenges, including the introduction of biases. These biases present immediate risks in the algorithms’ application. It was, for instance, shown that neural networks can deduce racial information solely from a patient’s X-ray scan, a task beyond the capability of medical experts. If this fact is not known to the medical expert, automatic decision-making based on this algorithm could lead to prescribing a treatment (purely) based on racial information. While current methodologies allow for the ‘‘orthogonalization’’ or ‘’normalization’’ of neural networks with respect to such information, existing approaches are grounded in linear models. Our paper advances the discourse by introducing corrections for non-linearities such as ReLU activations. Our approach also encompasses scalar and tensor-valued predictions, facilitating its integration into neural network architectures. Through extensive experiments, we validate our method’s effectiveness in safeguarding sensitive data in generalized linear models, normalizing convolutional neural networks for metadata, and rectifying pre-existing embeddings for undesired attributes.

A major challenge in sample-based inference (SBI) for Bayesian neural networks is the size and structure of the networks’ parameter space. Our work shows that successful SBI is possible by embracing the characteristic relationship between weight and function space, uncovering a systematic link between overparameterization and the difficulty of the sampling problem. Through extensive experiments, we establish practical guidelines for sampling and convergence diagnosis. As a result, we present a Bayesian deep ensemble approach as an effective solution with competitive performance and uncertainty quantification.

We show that utilizing attribution maps for training neural networks can improve regularization of models and thus increase performance. Regularization is key in deep learning, especially when training complex models on relatively small datasets. In order to understand inner workings of neural networks, attribution methods such as Layer-wise Relevance Propagation (LRP) have been extensively studied, particularly for interpreting the relevance of input features. We introduce Challenger, a module that leverages the explainable power of attribution maps in order to manipulate particularly relevant input patterns. Therefore, exposing and subsequently resolving regions of ambiguity towards separating classes on the ground-truth data manifold, an issue that arises particularly when training models on rather small datasets. Our Challenger module increases model performance through building more diverse filters within the network and can be applied to any input data domain. We demonstrate that our approach results in substantially better classification as well as calibration performance on datasets with only a few samples up to datasets with thousands of samples. In particular, we show that our generic domain-independent approach yields state-of-the-art results in vision, natural language processing and on time series tasks.

Counterfactual explanations elucidate algorithmic decisions by pointing to scenarios that would have led to an alternative, desired outcome. Giving insight into the model’s behavior, they hint users towards possible actions and give grounds for contesting decisions. As a crucial factor in achieving these goals, counterfactuals must be plausible, i.e., describing realistic alternative scenarios within the data manifold. This paper leverages a recently developed generative modeling technique – adversarial random forests (ARFs) – to efficiently generate plausible counterfactuals in a model-agnostic way. ARFs can serve as a plausibility measure or directly generate counterfactual explanations. Our ARF-based approach surpasses the limitations of existing methods that aim to generate plausible counterfactual explanations: It is easy to train and computationally highly efficient, handles continuous and categorical data naturally, and allows integrating additional desiderata such as sparsity in a straightforward manner.

While machine learning (ML) models are increasingly used due to their high predictive power, their use in understanding the data-generating process (DGP) is limited. Understanding the DGP requires insights into feature-target associations, which many ML models cannot directly provide due to their opaque internal mechanisms. Feature importance (FI) methods provide useful insights into the DGP under certain conditions. Since the results of different FI methods have different interpretations, selecting the correct FI method for a concrete use case is crucial and still requires expert knowledge. This paper serves as a comprehensive guide to help understand the different interpretations of global FI methods. Through an extensive review of FI methods and providing new proofs regarding their interpretation, we facilitate a thorough understanding of these methods and formulate concrete recommendations for scientific inference. We conclude by discussing options for FI uncertainty estimation and point to directions for future research aiming at full statistical inference from black-box ML models.

The recently developed Prior-Data Fitted Networks (PFNs) have shown very promising results for applications in low-data regimes. The TabPFN model, a special case of PFNs for tabular data, is able to achieve state-of-the-art performance on a variety of classification tasks while producing posterior predictive distributions in mere seconds by in-context learning without the need for learning parameters or hyperparameter tuning. This makes TabPFN a very attractive option for a wide range of domain applications. However, a major drawback of the method is its lack of interpretability. Therefore, we propose several adaptations of popular interpretability methods that we specifically design for TabPFN. By taking advantage of the unique properties of the model, our adaptations allow for more efficient computations than existing implementations. In particular, we show how in-context learning facilitates the estimation of Shapley values by avoiding approximate retraining and enables the use of Leave-One-Covariate-Out (LOCO) even when working with large-scale Transformers. In addition, we demonstrate how data valuation methods can be used to address scalability challenges of TabPFN.

Understanding how assignments of instances to clusters can be attributed to the features can be vital in many applications. However, research to provide such feature attributions has been limited. Clustering algorithms with built-in explanations are scarce. Common algorithm-agnostic approaches involve dimension reduction and subsequent visualization, which transforms the original features used to cluster the data; or training a supervised learning classifier on the found cluster labels, which adds additional and intractable complexity. We present FACT (feature attributions for clustering), an algorithm-agnostic framework that preserves the integrity of the data and does not introduce additional models. As the defining characteristic of FACT, we introduce a set of work stages: sampling, intervention, reassignment, and aggregation. Furthermore, we propose two novel FACT methods: SMART (scoring metric after permutation) measures changes in cluster assignments by custom scoring functions after permuting selected features; IDEA (isolated effect on assignment) indicates local and global changes in cluster assignments after making uniform changes to selected features.

This work introduces a novel R package for concise, informative summaries of machine learning models. We take inspiration from the summary function for (generalized) linear models in R, but extend it in several directions: First, our summary function is model-agnostic and provides a unified summary output also for non-parametric machine learning models; Second, the summary output is more extensive and customizable – it comprises information on the dataset, model performance, model complexity, model’s estimated feature importances, feature effects, and fairness metrics; Third, models are evaluated based on resampling strategies for unbiased estimates of model performances, feature importances, etc. Overall, the clear, structured output should help to enhance and expedite the model selection process, making it a helpful tool for practitioners and researchers alike.

Neural network representations of simple models, such as linear regression, are being studied increasingly to better understand the underlying principles of deep learning algorithms. However, neural representations of distributional regression models, such as the Cox model, have received little attention so far. We close this gap by proposing a framework for distributional regression using inverse flow transformations (DRIFT), which includes neural representations of the aforementioned models. We empirically demonstrate that the neural representations of models in DRIFT can serve as a substitute for their classical statistical counterparts in several applications involving continuous, ordered, time-series, and survival outcomes. We confirm that models in DRIFT empirically match the performance of several statistical methods in terms of estimation of partial effects, prediction, and aleatoric uncertainty quantification. DRIFT covers both interpretable statistical models and flexible neural networks opening up new avenues in both statistical modeling and deep learning.

We present a novel approach to uncertainty quantification in classification tasks based on label-wise decomposition of uncertainty measures. This label-wise perspective allows uncertainty to be quantified at the individual class level, thereby improving cost-sensitive decision-making and helping understand the sources of uncertainty. Furthermore, it allows to define total, aleatoric, and epistemic uncertainty on the basis of non-categorical measures such as variance, going beyond common entropy-based measures. In particular, variance-based measures address some of the limitations associated with established methods that have recently been discussed in the literature. We show that our proposed measures adhere to a number of desirable properties. Through empirical evaluation on a variety of benchmark data sets – including applications in the medical domain where accurate uncertainty quantification is crucial – we establish the effectiveness of label-wise uncertainty quantification.

Piecewise Exponential Additive Mixed Models (PAMMs) (Bender et al., 2018) have gained popularity in various domains due to their ability to tackle a wide variety of survival problems and their flexibility to model non-linear covariate effects, including time-varying effects and cumulative effects (Bender et al., 2019). One advantage of such reduction techniques is that they do not require any specialised software for the estimation of the model parameters. Thus, in the case of the PAMM, they can be conveniently estimated using generalized additive mixed modeling methodology or, for example, respective boosting or deep learning based approaches (Bender et al., 2022). Nevertheless, their use in practice requires pre-processing, which differs depending on the survival task at hand (e.g. left-truncation, competing risks, etc.) and post-processing (e.g. transforming estimated parameters to useful quantities like survival or transition probabilities). The R package pammtools facilitates the entire modeling process, so far, however, only for single-event data. Here we extend the framework and package capabilities to handle general multi-state models.

This work introduces a novel R package for concise, informative summaries of machine learning models. We take inspiration from the summary function for (generalized) linear models in R, but extend it in several directions: First, our summary function is model-agnostic and provides a unified summary output also for non-parametric machine learning models; Second, the summary output is more extensive and customizable – it comprises information on the dataset, model performance, model complexity, model’s estimated feature importances, feature effects, and fairness metrics; Third, models are evaluated based on resampling strategies for unbiased estimates of model performances, feature importances, etc. Overall, the clear, structured output should help to enhance and expedite the model selection process, making it a helpful tool for practitioners and researchers alike.

mlr3torch is a deep learning framework for the mlr3 ecosystem built on top of torch. It allows to easily build, train and evaluate deep learning models in a few lines of codes, without needing to worry about low-level details. Off-the-shelf learners are readily available, but custom architectures can be defined by connecting PipeOpTorch operators in an mlr3pipelines::Graph.

When different researchers study the same research question using the same dataset they may obtain different and potentially even conflicting results. This is because there is often substantial flexibility in researchers’ analytical choices, an issue also referred to as ‘‘researcher degrees of freedom’’. Combined with selective reporting of the smallest p-value or largest effect, researcher degrees of freedom may lead to an increased rate of false positive and overoptimistic results. In this paper, we address this issue by formalizing the multiplicity of analysis strategies as a multiple testing problem. As the test statistics of different analysis strategies are usually highly dependent, a naive approach such as the Bonferroni correction is inappropriate because it leads to an unacceptable loss of power. Instead, we propose using the ‘‘minP’’ adjustment method, which takes potential test dependencies into account and approximates the underlying null distribution of the minimal p-value through a permutation-based procedure. This procedure is known to achieve more power than simpler approaches while ensuring a weak control of the family-wise error rate. We illustrate our approach for addressing researcher degrees of freedom by applying it to a study on the impact of perioperative paO2 on post-operative complications after neurosurgery. A total of 48 analysis strategies are considered and adjusted using the minP procedure. This approach allows to selectively report the result of the analysis strategy yielding the most convincing evidence, while controlling the type 1 error – and thus the risk of publishing false positive results that may not be replicable.