We propose a function-valued evaluation metric for generative models based on the relative density ratio (RDR) designed to characterize distributional differences between real and generated samples. As an evaluation metric, the RDR function preserves $ϕ$-divergence between two distributions, enables sample-level evaluation that facilitates downstream investigations of feature-specific distributional differences, and has a bounded range that affords clear interpretability and numerical stability. Function estimation of the RDR is achieved efficiently through optimization on the variational form of $ϕ$-divergence. We provide theoretical convergence rate guarantees for general estimators based on M-estimator theory, as well as the convergence rate of neural network-based estimators when the true ratio is in the anisotropic Besov space. We demonstrate the power of the proposed RDR-based evaluation through numerical experiments on MNIST, CelebA64, and the American Gut project microbiome data. We show that the estimated RDR enables not only effective overall comparison of competing generative models, but also a convenient way to reveal the underlying nature of goodness-of-fit. This enables one to assess support overlap, coverage, and fidelity while pinpointing regions of the sample space where generators concentrate and revealing the features that drive the most salient distributional differences.

Exploring large corpora can generate useful discoveries but is time-consuming for humans. We formulate a new task, D5, that automatically discovers differences between two large corpora in a goal-driven way. The task input is a problem comprising a user-specified research goal (" ") and a corpus pair (two large collections of patients' self-reported reactions after taking each drug). The output is a goal-related description (discovery) of how these corpora differ (patients taking drug A " " more often). We build a D5 system, and to quantitatively evaluate its performance, we 1) build a diagnostic benchmark, SynD5, to test whether it can recover known differences between two synthetic corpora, and 2) contribute a meta-dataset, OpenD5, aggregating 675 open-ended problems ranging across business, social sciences, humanities, machine learning, and health. With both synthetic and real datasets, we confirm that language models can leverage the user-specified goals to propose more relevant candidate discoveries, and they sometimes produce discoveries previously unknown to the authors, including demographic differences in discussion topics, political stances in speech, insights in commercial reviews, and error patterns in NLP models. Finally, we discuss the limitations of the current D5 system, which discovers correlation rather than causation and has the potential to reinforce societal biases when misused; therefore, practitioners should treat the outputs of our system with caution.

In distributed and federated learning, heterogeneity across data sources remains a major obstacle to effective model aggregation and convergence. We focus on feature heterogeneity and introduce energy distance as a sensitive measure for quantifying distributional discrepancies. While we show that energy distance is robust for detecting data distribution shifts, its direct use in large-scale systems can be prohibitively expensive. To address this, we develop Taylor approximations that preserve key theoretical quantitative properties while reducing computational overhead. Through simulation studies, we show how accurately capturing feature discrepancies boosts convergence in distributed learning. Finally, we propose a novel application of energy distance to assign penalty weights for aligning predictions across heterogeneous nodes, ultimately enhancing coordination in federated and distributed settings.

Exploring large corpora can generate useful discoveries but is time-consuming for humans. We formulate a new task, D5, that automatically discovers differences between two large corpora in a goal-driven way. The task input is a problem comprising a user-specified research goal ("comparing the side effects of drug A and drug") and a corpus pair (two large collections of patients' self-reported reactions after taking each drug). The output is a goal-related description (discovery) of how these corpora differ (patients taking drug A "mention feelings of paranoia" more often). We build a D5 system, and to quantitatively evaluate its performance, we 1) build a diagnostic benchmark, SynD5, to test whether it can recover known differences between two synthetic corpora, and 2) contribute a meta-dataset, OpenD5, aggregating 675 open-ended problems ranging across business, social sciences, humanities, machine learning, and health.

The development of causal prediction models is challenged by the fact that the outcome is only observable for the applied (factual) intervention and not for its alternatives (the so-called counterfactuals); in medicine we only know patients' survival for the administered drug and not for other therapeutic options. Machine learning approaches for counterfactual reasoning have to deal with both unobserved outcomes and distributional differences due to non-random treatment administration. Unsupervised domain adaptation (UDA) addresses similar issues; one has to deal with unobserved outcomes -- the labels of the target domain -- and distributional differences between source and target domain. We propose Adversarial Distribution Balancing for Counterfactual Reasoning (ADBCR), which directly uses potential outcome estimates of the counterfactuals to remove spurious causal relations. We show that ADBCR outcompetes state-of-the-art methods on three benchmark datasets, and demonstrate that ADBCR's performance can be further improved if unlabeled validation data are included in the training procedure to better adapt the model to the validation domain.

Mining large corpora can generate useful discoveries but is time-consuming for humans. We formulate a new task, D5, that automatically discovers differences between two large corpora in a goal-driven way. The task input is a problem comprising a research goal "$\textit{comparing the side effects of drug A and drug B}$" and a corpus pair (two large collections of patients' self-reported reactions after taking each drug). The output is a language description (discovery) of how these corpora differ (patients taking drug A "$\textit{mention feelings of paranoia}$" more often). We build a D5 system, and to quantitatively measure its performance, we 1) contribute a meta-dataset, OpenD5, aggregating 675 open-ended problems ranging across business, social sciences, humanities, machine learning, and health, and 2) propose a set of unified evaluation metrics: validity, relevance, novelty, and significance. With the dataset and the unified metrics, we confirm that language models can use the goals to propose more relevant, novel, and significant candidate discoveries. Finally, our system produces discoveries previously unknown to the authors on a wide range of applications in OpenD5, including temporal and demographic differences in discussion topics, political stances and stereotypes in speech, insights in commercial reviews, and error patterns in NLP models.

Visual recognition models are prone to learning spurious correlations induced by a biased training set where certain conditions $B$ (\eg, Indoors) are over-represented in certain classes $Y$ (\eg, Big Dogs). Synthetic data from generative models offers a promising direction to mitigate this issue by augmenting underrepresented conditions in the real dataset. However, this introduces another potential source of bias from generative model artifacts in the synthetic data. Indeed, as we will show, prior work uses synthetic data to resolve the model's bias toward $B$, but it doesn't correct the models' bias toward the pair $(B, G)$ where $G$ denotes whether the sample is real or synthetic. Thus, the model could simply learn signals based on the pair $(B, G)$ (\eg, Synthetic Indoors) to make predictions about $Y$ (\eg, Big Dogs). To address this issue, we propose a two-step training pipeline that we call From Fake to Real (FFR). The first step of FFR pre-trains a model on balanced synthetic data to learn robust representations across subgroups. In the second step, FFR fine-tunes the model on real data using ERM or common loss-based bias mitigation methods. By training on real and synthetic data separately, FFR avoids the issue of bias toward signals from the pair $(B, G)$. In other words, synthetic data in the first step provides effective unbiased representations that boosts performance in the second step. Indeed, our analysis of high bias setting (99.9\%) shows that FFR improves performance over the state-of-the-art by 7-14\% over three datasets (CelebA, UTK-Face, and SpuCO Animals).

Modern machine learning approaches excel in static settings where a large amount of i.i.d. training data are available for a given task. In a dynamic environment, though, an intelligent agent needs to be able to transfer knowledge and re-use learned components across domains. It has been argued that this may be possible through causal models, aiming to mirror the modularity of the real world in terms of independent causal mechanisms. However, the true causal structure underlying a given set of data is generally not identifiable, so it is desirable to have means to quantify differences between models (e.g., between the ground truth and an estimate), on both the observational and interventional level. In the present work, we introduce the Interventional Kullback-Leibler (IKL) divergence to quantify both structural and distributional differences between models based on a finite set of multi-environment distributions generated by interventions from the ground truth. Since we generally cannot quantify all differences between causal models for every finite set of interventional distributions, we propose a sufficient condition on the intervention targets to identify subsets of observed variables on which the models provably agree or disagree.

As machine learning (ML) models gain traction in clinical applications, understanding the impact of clinician and societal biases on ML models is increasingly important. While biases can arise in the labels used for model training, the many sources from which these biases arise are not yet well-studied. In this paper, we highlight disparate censorship (i.e., differences in testing rates across patient groups) as a source of label bias that clinical ML models may amplify, potentially causing harm. Many patient risk-stratification models are trained using the results of clinician-ordered diagnostic and laboratory tests of labels. Patients without test results are often assigned a negative label, which assumes that untested patients do not experience the outcome. Since orders are affected by clinical and resource considerations, testing may not be uniform in patient populations, giving rise to disparate censorship. Disparate censorship in patients of equivalent risk leads to undertesting in certain groups, and in turn, more biased labels for such groups. Using such biased labels in standard ML pipelines could contribute to gaps in model performance across patient groups. Here, we theoretically and empirically characterize conditions in which disparate censorship or undertesting affect model performance across subgroups. Our findings call attention to disparate censorship as a source of label bias in clinical ML models.

In this paper, we propose a new framework to detect adversarial examples motivated by the observations that random components can improve the smoothness of predictors and make it easier to simulate output distribution of deep neural network. With these observations, we propose a novel Bayesian adversarial example detector, short for BATer, to improve the performance of adversarial example detection. In specific, we study the distributional difference of hidden layer output between natural and adversarial examples, and propose to use the randomness of Bayesian neural network (BNN) to simulate hidden layer output distribution and leverage the distribution dispersion to detect adversarial examples. The advantage of BNN is that the output is stochastic while neural networks without random components do not have such characteristics. Empirical results on several benchmark datasets against popular attacks show that the proposed BATer outperforms the state-of-the-art detectors in adversarial example detection.