Bayesian deep learning (BDL) is a promising approach to achieve well-calibrated predictions on distribution-shifted data. Nevertheless, there exists no large-scale survey that evaluates recent SOTA methods on diverse, realistic, and challenging benchmark tasks in a systematic manner. To provide a clear picture of the current state of BDL research, we evaluate modern BDL algorithms on real-world datasets from the WILDS collection containing challenging classification and regression tasks, with a focus on generalization capability and calibration under distribution shift. We compare the algorithms on a wide range of large, convolutional and transformer-based neural network architectures. In particular, we investigate a signed version of the expected calibration error that reveals whether the methods are over-or underconfident, providing further insight into the behavior of the methods. Further, we provide the first systematic evaluation of BDL for fine-tuning large pre-trained models, where training from scratch is prohibitively expensive. Finally, given the recent success of Deep Ensembles, we extend popular single-mode posterior approximations to multiple modes by the use of ensembles. While we find that ensembling single-mode approximations generally improves the generalization capability and calibration of the models by a significant margin, we also identify a failure mode of ensembles when finetuning large transformer-based language models. In this setting, variational inference based approaches such as last-layer Bayes By Backprop outperform other methods in terms of accuracy by a large margin, while modern approximate inference algorithms such as SWAG achieve the best calibration.

Bayesian deep learning (BDL) is a promising approach to achieve well-calibrated predictions on distribution-shifted data. Nevertheless, there exists no large-scale survey that evaluates recent SOTA methods on diverse, realistic, and challenging benchmark tasks in a systematic manner. To provide a clear picture of the current state of BDL research, we evaluate modern BDL algorithms on real-world datasets from the WILDS collection containing challenging classification and regression tasks, with a focus on generalization capability and calibration under distribution shift. We compare the algorithms on a wide range of large, convolutional and transformer-based neural network architectures. In particular, we investigate a signed version of the expected calibration error that reveals whether the methods are over- or underconfident, providing further insight into the behavior of the methods.

As deep learning models are increasingly deployed in safety-critical applications, evaluating their vulnerabilities to adversarial perturbations is essential for ensuring their reliability and trustworthiness. Over the past decade, a large number of white-box adversarial robustness evaluation methods (i.e., attacks) have been proposed, ranging from single-step to multi-step methods and from individual to ensemble methods. Despite these advances, challenges remain in conducting meaningful and comprehensive robustness evaluations, particularly when it comes to large-scale testing and ensuring evaluations reflect real-world adversarial risks. In this work, we focus on image classification models and propose a novel individual attack method, Probability Margin Attack (PMA), which defines the adversarial margin in the probability space rather than the logits space. We analyze the relationship between PMA and existing cross-entropy or logits-margin-based attacks, and show that PMA can outperform the current state-of-the-art individual methods. Building on PMA, we propose two types of ensemble attacks that balance effectiveness and efficiency. Furthermore, we create a million-scale dataset, CC1M, derived from the existing CC3M dataset, and use it to conduct the first million-scale white-box adversarial robustness evaluation of adversarially-trained ImageNet models. Our findings provide valuable insights into the robustness gaps between individual versus ensemble attacks and small-scale versus million-scale evaluations.

In display advertising, advertisers want to achieve a marketing objective with constraints on budget and cost-per-outcome. This is usually formulated as an optimization problem that maximizes the total utility under constraints. The optimization is carried out in an online fashion in the dual space - for an incoming Ad auction, a bid is placed using an optimal bidding formula, assuming optimal values for the dual variables; based on the outcome of the previous auctions, the dual variables are updated in an online fashion. While this approach is theoretically sound, in practice, the dual variables are not optimal from the beginning, but rather converge over time. Specifically, for the cost-constraint, the convergence is asymptotic. As a result, we find that cost-control is ineffective. In this work, we analyse the shortcomings of the optimal bidding formula and propose a modification that deviates from the theoretical derivation. We simulate various practical scenarios and study the cost-control behaviors of the two algorithms. Through a large-scale evaluation on the real-word data, we show that the proposed modification reduces the cost violations by 50%, thereby achieving a better cost-control than the theoretical bidding formula.

Monte Carlo methods, Variational Inference, and their combinations play a pivotal role in sampling from intractable probability distributions. However, current studies lack a unified evaluation framework, relying on disparate performance measures and limited method comparisons across diverse tasks, complicating the assessment of progress and hindering the decision-making of practitioners. In response to these challenges, our work introduces a benchmark that evaluates sampling methods using a standardized task suite and a broad range of performance criteria. Moreover, we study existing metrics for quantifying mode collapse and introduce novel metrics for this purpose. Our findings provide insights into strengths and weaknesses of existing sampling methods, serving as a valuable reference for future developments. The code is publicly available here.

Bayesian deep learning (BDL) is a promising approach to achieve well-calibrated predictions on distribution-shifted data. Nevertheless, there exists no large-scale survey that evaluates recent SOTA methods on diverse, realistic, and challenging benchmark tasks in a systematic manner. To provide a clear picture of the current state of BDL research, we evaluate modern BDL algorithms on real-world datasets from the WILDS collection containing challenging classification and regression tasks, with a focus on generalization capability and calibration under distribution shift. We compare the algorithms on a wide range of large, convolutional and transformer-based neural network architectures. In particular, we investigate a signed version of the expected calibration error that reveals whether the methods are over- or under-confident, providing further insight into the behavior of the methods. Further, we provide the first systematic evaluation of BDL for fine-tuning large pre-trained models, where training from scratch is prohibitively expensive. Finally, given the recent success of Deep Ensembles, we extend popular single-mode posterior approximations to multiple modes by the use of ensembles. While we find that ensembling single-mode approximations generally improves the generalization capability and calibration of the models by a significant margin, we also identify a failure mode of ensembles when finetuning large transformer-based language models. In this setting, variational inference based approaches such as last-layer Bayes By Backprop outperform other methods in terms of accuracy by a large margin, while modern approximate inference algorithms such as SWAG achieve the best calibration.

Topic models are a class of unsupervised learning algorithms for detecting the semantic structure within a text corpus. Together with a subsequent dimensionality reduction algorithm, topic models can be used for deriving spatializations for text corpora as two-dimensional scatter plots, reflecting semantic similarity between the documents and supporting corpus analysis. Although the choice of the topic model, the dimensionality reduction, and their underlying hyperparameters significantly impact the resulting layout, it is unknown which particular combinations result in high-quality layouts with respect to accuracy and perception metrics. To investigate the effectiveness of topic models and dimensionality reduction methods for the spatialization of corpora as two-dimensional scatter plots (or basis for landscape-type visualizations), we present a large-scale, benchmark-based computational evaluation. Our evaluation consists of (1) a set of corpora, (2) a set of layout algorithms that are combinations of topic models and dimensionality reductions, and (3) quality metrics for quantifying the resulting layout. The corpora are given as document-term matrices, and each document is assigned to a thematic class. The chosen metrics quantify the preservation of local and global properties and the perceptual effectiveness of the two-dimensional scatter plots. By evaluating the benchmark on a computing cluster, we derived a multivariate dataset with over 45 000 individual layouts and corresponding quality metrics. Based on the results, we propose guidelines for the effective design of text spatializations that are based on topic models and dimensionality reductions. As a main result, we show that interpretable topic models are beneficial for capturing the structure of text corpora. We furthermore recommend the use of t-SNE as a subsequent dimensionality reduction.

Recently introduced transformer-based article encoders (TAEs) designed to produce similar vector representations for mutually related scientific articles have demonstrated strong performance on benchmark datasets for scientific article recommendation. However, the existing benchmark datasets are predominantly focused on single domains and, in some cases, contain easy negatives in small candidate pools. Evaluating representations on such benchmarks might obscure the realistic performance of TAEs in setups with thousands of articles in candidate pools. In this work, we evaluate TAEs on large benchmarks with more challenging candidate pools. We compare the performance of TAEs with a lexical retrieval baseline model BM25 on the task of citation recommendation, where the model produces a list of recommendations for citing in a given input article. We find out that BM25 is still very competitive with the state-of-the-art neural retrievers, a finding which is surprising given the strong performance of TAEs on small benchmarks. As a remedy for the limitations of the existing benchmarks, we propose a new benchmark dataset for evaluating scientific article representations: Multi-Domain Citation Recommendation dataset (MDCR), which covers different scientific fields and contains challenging candidate pools.

EEG is the most common signal source for noninvasive BCI applications. For such applications, the EEG signal needs to be decoded and translated into appropriate actions. A recently emerging EEG decoding approach is deep learning with Convolutional or Recurrent Neural Networks (CNNs, RNNs) with many different architectures already published. Here we present a novel framework for the large-scale evaluation of different deep-learning architectures on different EEG datasets. This framework comprises (i) a collection of EEG datasets currently comprising 100 examples (recording sessions) from six different classification problems, (ii) a collection of different EEG decoding algorithms, and (iii) a wrapper linking the decoders to the data as well as handling structured documentation of all settings and (hyper-) parameters and statistics, designed to ensure transparency and reproducibility. As an applications example we used our framework by comparing three publicly available CNN architectures: the Braindecode Deep4 ConvNet, Braindecode Shallow ConvNet, and EEGNet. We also show how our framework can be used to study similarities and differences in the performance of different decoding methods across tasks. We argue that the deep learning EEG framework as described here could help to tap the full potential of deep learning for BCI applications.