In today's data-driven era, machine learning models are designed to reach higher performance, and the size of new models will then inherently increase, therefore the information stored (or memorized) in the parameters [Hartley and Tsaftaris, 2022, Del Grosso et al., 2023]. The protection of sensitive information is of paramount importance. Membership Inference Attacks (MIAs) have emerged as a concerning threat, capable of unveiling whether a specific data point was part of the training dataset of a machine learning model [Shokri et al., 2017, Nasr et al., 2019, Song et al., 2017a, Zhu et al., 2019]. Such attacks can potentially compromise individual privacy and security by exposing sensitive information [Carlini et al., 2023a]. Furthermore, a recent publication [Tabassi et al., 2019] from the National Institute of Standards and Technology (NIST) explicitly notes that an MIA that successfully identifies an individual as part of the dataset used for training the target model constitutes a breach of confidentiality. To date, the most comprehensive defense mechanism against privacy attacks is differential privacy (DP), a framework initially introduced by Dwork et al. [2006]. DP has shown remarkable adaptability in safeguarding the privacy of machine learning models during training, as demonstrated by the works of Hannun et al. [2021], Jayaraman and Evans [2019]. However, it is worth noting that achieving a high level of privacy through differentially private training often comes at a significant cost to the accuracy of the model, especially when aiming for a low privacy parameter [Sablayrolles et al., 2019]. Conversely, when evaluating the practical effectiveness of DP in terms of its ability to protect against privacy attacks empirically, the outlook is considerably more positive.

Uncertainty Quantification (UQ) is essential for creating trustworthy machine learning models. Recent years have seen a steep rise in UQ methods that can flag suspicious examples, however, it is often unclear what exactly these methods identify. In this work, we propose a framework for categorizing uncertain examples flagged by UQ methods in classification tasks. We introduce the confusion density matrix -- a kernel-based approximation of the misclassification density -- and use this to categorize suspicious examples identified by a given uncertainty method into three classes: out-of-distribution (OOD) examples, boundary (Bnd) examples, and examples in regions of high in-distribution misclassification (IDM). Through extensive experiments, we show that our framework provides a new and distinct perspective for assessing differences between uncertainty quantification methods, thereby forming a valuable assessment benchmark.

We consider the problem of estimating the mean of a random variable Y subject to non-ignorable missingness, i.e., where the missingness mechanism depends on Y . We connect the auxiliary proxy variable framework for non-ignorable missingness (West and Little, 2013) to the label shift setting (Saerens et al., 2002). Exploiting this connection, we construct an estimator for non-ignorable missing data that uses high-dimensional covariates (or proxies) without the need for a generative model. In synthetic and semi-synthetic experiments, we study the behavior of the proposed estimator, comparing it to commonly used ignorable estimators in both well-specified and misspecified settings. Additionally, we develop a score to assess how consistent the data are with the label shift assumption. We use our approach to estimate disease prevalence using a large health survey, comparing ignorable and non-ignorable approaches. We show that failing to account for non-ignorable missingness can have profound consequences on conclusions drawn from non-representative samples.

Scientists continue to develop increasingly complex mechanistic models to reflect their knowledge more realistically. Statistical inference using these models can be challenging since the corresponding likelihood function is often intractable and model simulation may be computationally burdensome. Fortunately, in many of these situations, it is possible to adopt a surrogate model or approximate likelihood function. It may be convenient to conduct Bayesian inference directly with the surrogate, but this can result in bias and poor uncertainty quantification. In this paper we propose a new method for adjusting approximate posterior samples to reduce bias and produce more accurate uncertainty quantification. We do this by optimizing a transform of the approximate posterior that maximizes a scoring rule. Our approach requires only a (fixed) small number of complex model simulations and is numerically stable. We demonstrate good performance of the new method on several examples of increasing complexity.

This manuscript introduces a new optimization framework for machine learning and AI, named {\bf empirical X-risk minimization (EXM)}. X-risk is a term introduced to represent a family of compositional measures or objectives, in which each data point is compared with a large number of items explicitly or implicitly for defining a risk function. It includes surrogate objectives of many widely used measures and non-decomposable losses, e.g., AUROC, AUPRC, partial AUROC, NDCG, MAP, precision/recall at top $K$ positions, precision at a certain recall level, listwise losses, p-norm push, top push, global contrastive losses, etc. While these non-decomposable objectives and their optimization algorithms have been studied in the literature of machine learning, computer vision, information retrieval, and etc, optimizing these objectives has encountered some unique challenges for deep learning. In this paper, we present recent rigorous efforts for EXM with a focus on its algorithmic foundations and its applications. We introduce a class of algorithmic techniques for solving EXM with smooth non-convex objectives. We formulate EXM into three special families of non-convex optimization problems belonging to non-convex compositional optimization, non-convex min-max optimization and non-convex bilevel optimization, respectively. For each family of problems, we present some strong baseline algorithms and their complexities, which will motivate further research for improving the existing results. Discussions about the presented results and future studies are given at the end. Efficient algorithms for optimizing a variety of X-risks are implemented in the LibAUC library at \url{www.libauc.org}.

Empirical risk minimization (ERM) is sensitive to spurious correlations in the training data, which poses a significant risk when deploying systems trained under this paradigm in high-stake applications. While the existing literature focuses on maximizing group-balanced or worst-group accuracy, estimating these accuracies is hindered by costly bias annotations. This study contends that current bias-unsupervised approaches to group robustness continue to rely on group information to achieve optimal performance. Firstly, these methods implicitly assume that all group combinations are represented during training. To illustrate this, we introduce a systematic generalization task on the MPI3D dataset and discover that current algorithms fail to improve the ERM baseline when combinations of observed attribute values are missing. Secondly, bias labels are still crucial for effective model selection, restricting the practicality of these methods in real-world scenarios. To address these limitations, we propose a revised methodology for training and validating debiased models in an entirely bias-unsupervised manner. We achieve this by employing pretrained self-supervised models to reliably extract bias information, which enables the integration of a logit adjustment training loss with our validation criterion. Our empirical analysis on synthetic and real-world tasks provides evidence that our approach overcomes the identified challenges and consistently enhances robust accuracy, attaining performance which is competitive with or outperforms that of state-of-the-art methods, which, conversely, rely on bias labels for validation.

Joint feature selection and tree ensemble learning is a challenging task. Popular tree ensemble toolkits e.g., Gradient Boosted Trees and Random Forests support feature selection post-training based on feature importances, which are known to be misleading, and can significantly hurt performance. We propose Skinny Trees: a toolkit for feature selection in tree ensembles, such that feature selection and tree ensemble learning occurs simultaneously. It is based on an end-to-end optimization approach that considers feature selection in differentiable trees with Group $\ell_0 - \ell_2$ regularization. We optimize with a first-order proximal method and present convergence guarantees for a non-convex and non-smooth objective. Interestingly, dense-to-sparse regularization scheduling can lead to more expressive and sparser tree ensembles than vanilla proximal method. On 15 synthetic and real-world datasets, Skinny Trees can achieve $1.5\times$ - $620\times$ feature compression rates, leading up to $10\times$ faster inference over dense trees, without any loss in performance. Skinny Trees lead to superior feature selection than many existing toolkits e.g., in terms of AUC performance for $25\%$ feature budget, Skinny Trees outperforms LightGBM by $10.2\%$ (up to $37.7\%$), and Random Forests by $3\%$ (up to $12.5\%$).

We introduce contextual stochastic bilevel optimization (CSBO) -- a stochastic bilevel optimization framework with the lower-level problem minimizing an expectation conditioned on some contextual information and the upper-level decision variable. This framework extends classical stochastic bilevel optimization when the lower-level decision maker responds optimally not only to the decision of the upper-level decision maker but also to some side information and when there are multiple or even infinite many followers. It captures important applications such as meta-learning, personalized federated learning, end-to-end learning, and Wasserstein distributionally robust optimization with side information (WDRO-SI). Due to the presence of contextual information, existing single-loop methods for classical stochastic bilevel optimization are unable to converge. To overcome this challenge, we introduce an efficient double-loop gradient method based on the Multilevel Monte-Carlo (MLMC) technique and establish its sample and computational complexities. When specialized to stochastic nonconvex optimization, our method matches existing lower bounds. For meta-learning, the complexity of our method does not depend on the number of tasks. Numerical experiments further validate our theoretical results.

Contrastive analysis (CA) refers to the exploration of variations uniquely enriched in a target dataset as compared to a corresponding background dataset generated from sources of variation that are irrelevant to a given task. For example, a biomedical data analyst may wish to find a small set of genes to use as a proxy for variations in genomic data only present among patients with a given disease (target) as opposed to healthy control subjects (background). However, as of yet the problem of feature selection in the CA setting has received little attention from the machine learning community. In this work we present contrastive feature selection (CFS), a method for performing feature selection in the CA setting. We motivate our approach with a novel information-theoretic analysis of representation learning in the CA setting, and we empirically validate CFS on a semi-synthetic dataset and four real-world biomedical datasets. We find that our method consistently outperforms previously proposed state-of-the-art supervised and fully unsupervised feature selection methods not designed for the CA setting. An open-source implementation of our method is available at https://github.com/suinleelab/CFS.

We propose a general class of sample based explanations of machine learning models, which we term generalized representers. To measure the effect of a training sample on a model's test prediction, generalized representers use two components: a global sample importance that quantifies the importance of the training point to the model and is invariant to test samples, and a local sample importance that measures similarity between the training sample and the test point with a kernel. A key contribution of the paper is to show that generalized representers are the only class of sample based explanations satisfying a natural set of axiomatic properties. We discuss approaches to extract global importances given a kernel, and also natural choices of kernels given modern non-linear models. As we show, many popular existing sample based explanations could be cast as generalized representers with particular choices of kernels and approaches to extract global importances. Additionally, we conduct empirical comparisons of different generalized representers on two image and two text classification datasets.