Towards Poisoning Fair Representations

Fair machine learning seeks to mitigate model prediction bias against certain demographic subgroups such as elder and female. Recently, fair representation learning (FRL) trained by deep neural networks has demonstrated superior performance, whereby representations containing no demographic information are inferred from the data and then used as the input to classification or other downstream tasks. Despite the development of FRL methods, their vulnerability under data poisoning attack, a popular protocol to benchmark model robustness under adversarial scenarios, is under-explored. Data poisoning attacks have been developed for classical fair machine learning methods which incorporate fairness constraints into shallow-model classifiers. Nonetheless, these attacks fall short in FRL due to notably different fairness goals and model architectures. This work proposes the first data poisoning framework attacking FRL. We induce the model to output unfair representations that contain as much demographic information as possible by injecting carefully crafted poisoning samples into the training data. This attack entails a prohibitive bilevel optimization, wherefore an effective approximated solution is proposed. A theoretical analysis on the needed number of poisoning samples is derived and sheds light on defending against the attack. Experiments on benchmark fairness datasets and state-of-the-art fair representation learning models demonstrate the superiority of our attack.

SimFair: A Unified Framework for Fairness-Aware Multi-Label Classification

Recent years have witnessed increasing concerns towards unfair decisions made by machine learning algorithms. To improve fairness in model decisions, various fairness notions have been proposed and many fairness-aware methods are developed. However, most of existing definitions and methods focus only on single-label classification. Fairness for multi-label classification, where each instance is associated with more than one labels, is still yet to establish. To fill this gap, we study fairness-aware multi-label classification in this paper. We start by extending Demographic Parity (DP) and Equalized Opportunity (EOp), two popular fairness notions, to multi-label classification scenarios. Through a systematic study, we show that on multi-label data, because of unevenly distributed labels, EOp usually fails to construct a reliable estimate on labels with few instances. We then propose a new framework named Similarity $s$-induced Fairness ($s_\gamma$-SimFair). This new framework utilizes data that have similar labels when estimating fairness on a particular label group for better stability, and can unify DP and EOp. Theoretical analysis and experimental results on real-world datasets together demonstrate the advantage of over existing methods $s_\gamma$-SimFair on multi-label classification tasks.

COEP: Cascade Optimization for Inverse Problems with Entropy-Preserving Hyperparameter Tuning

We propose COEP, an automated and principled framework to solve inverse problems with deep generative models. COEP consists of two components, a cascade algorithm for optimization and an entropy-preserving criterion for hyperparameter tuning. Through COEP, the two components build up an efficient and end-to-end solver for inverse problems that require no human evaluation. We establish theoretical guarantees for the proposed methods. We also empirically validate the strength of COEP on denoising and noisy compressed sensing, which are two fundamental tasks in inverse problems.

Density Regression and Uncertainty Quantification with Bayesian Deep Noise Neural Networks

Deep neural network (DNN) models have achieved state-of-the-art predictive accuracy in a wide range of supervised learning applications. However, accurately quantifying the uncertainty in DNN predictions remains a challenging task. For continuous outcome variables, an even more difficult problem is to estimate the predictive density function, which not only provides a natural quantification of the predictive uncertainty, but also fully captures the random variation in the outcome. In this work, we propose the Bayesian Deep Noise Neural Network (B-DeepNoise), which generalizes standard Bayesian DNNs by extending the random noise variable from the output layer to all hidden layers. The latent random noise equips B-DeepNoise with the flexibility to approximate highly complex predictive distributions and accurately quantify predictive uncertainty. For posterior computation, the unique structure of B-DeepNoise leads to a closed-form Gibbs sampling algorithm that iteratively simulates from the posterior full conditional distributions of the model parameters, circumventing computationally intensive Metropolis-Hastings methods. A theoretical analysis of B-DeepNoise establishes a recursive representation of the predictive distribution and decomposes the predictive variance with respect to the latent parameters. We evaluate B-DeepNoise against existing methods on benchmark regression datasets, demonstrating its superior performance in terms of prediction accuracy, uncertainty quantification accuracy, and uncertainty quantification efficiency. To illustrate our method's usefulness in scientific studies, we apply B-DeepNoise to predict general intelligence from neuroimaging features in the Adolescent Brain Cognitive Development (ABCD) project.

Flows Succeed Where GANs Fail: Lessons from Low-Dimensional Data

Normalizing flows and generative adversarial networks (GANs) are both approaches to density estimation that use deep neural networks to transform samples from an uninformative prior distribution to an approximation of the data distribution. There is great interest in both for general-purpose statistical modeling, but the two approaches have seldom been compared to each other for modeling non-image data. The difficulty of computing likelihoods with GANs, which are implicit models, makes conducting such a comparison challenging. We work around this difficulty by considering several low-dimensional synthetic datasets. An extensive grid search over GAN architectures, hyperparameters, and training procedures suggests that no GAN is capable of modeling our simple low-dimensional data well, a task we view as a prerequisite for an approach to be considered suitable for general-purpose statistical modeling. Several normalizing flows, on the other hand, excelled at these tasks, even substantially outperforming WGAN in terms of Wasserstein distance---the metric that WGAN alone targets. Overall, normalizing flows appear to be more reliable tools for statistical inference than GANs.

Dimensionality Reduction on Grassmannian via Riemannian Optimization: A Generalized Perspective

This paper proposes a generalized framework with joint normalization which learns lower-dimensional subspaces with maximum discriminative power by making use of the Riemannian geometry. In particular, we model the similarity/dissimilarity between subspaces using various metrics defined on Grassmannian and formulate dimen-sionality reduction as a non-linear constraint optimization problem considering the orthogonalization. To obtain the linear mapping, we derive the components required to per-form Riemannian optimization (e.g., Riemannian conju-gate gradient) from the original Grassmannian through an orthonormal projection. We respect the Riemannian ge-ometry of the Grassmann manifold and search for this projection directly from one Grassmann manifold to an-other face-to-face without any additional transformations. In this natural geometry-aware way, any metric on the Grassmann manifold can be resided in our model theoreti-cally. We have combined five metrics with our model and the learning process can be treated as an unconstrained optimization problem on a Grassmann manifold. Exper-iments on several datasets demonstrate that our approach leads to a significant accuracy gain over state-of-the-art methods.