Andrii Kliachkin

Adam Bosák, Andrii Kliachkin, Jana Lepšová et al.

Constrained machine learning enables fairness-aware training, physics-informed neural networks, and integration of symbolic domain knowledge into statistical models. Despite its practical importance, no general method exists for the non-convex, non-smooth, stochastic setting that arises naturally in deep learning. We propose the Stochastic Penalty-Barrier Method (SPBM), which extends classical penalty and barrier methods to this setting via exponential dual averaging, a~stabilized penalty schedule, and the Moreau envelope to handle non-smoothness. Experiments across multiple settings show that SPBM matches or outperforms existing constrained optimization baselines while incurring only linear runtime overhead compared to unconstrained Adam for up to 10,000 constraints.

Andrii Kliachkin, Jana Lepšová, Gilles Bareilles et al.

There has been a considerable interest in constrained training of deep neural networks (DNNs) recently for applications such as fairness and safety. Several toolkits have been proposed for this task, yet there is still no industry standard. We present humancompatible.train (https://github.com/humancompatible/train), an easily-extendable PyTorch-based Python package for training DNNs with stochastic constraints. We implement multiple previously unimplemented algorithms for stochastically constrained stochastic optimization. We demonstrate the toolkit use by comparing two algorithms on a deep learning task with fairness constraints.

Andrii Kliachkin, Jana Lepšová, Gilles Bareilles et al.

The ability to train Deep Neural Networks (DNNs) with constraints is instrumental in improving the fairness of modern machine-learning models. Many algorithms have been analysed in recent years, and yet there is no standard, widely accepted method for the constrained training of DNNs. In this paper, we provide a challenging benchmark of real-world large-scale fairness-constrained learning tasks, built on top of the US Census (Folktables). We point out the theoretical challenges of such tasks and review the main approaches in stochastic approximation algorithms. Finally, we demonstrate the use of the benchmark by implementing and comparing three recently proposed, but as-of-yet unimplemented, algorithms both in terms of optimization performance, and fairness improvement. We release the code of the benchmark as a Python package at https://github.com/humancompatible/train.

Andrii Kliachkin, Eleni Psaroudaki, Jakub Marecek et al.

There has been great interest in fairness in machine learning, especially in relation to classification problems. In ranking-related problems, such as in online advertising, recommender systems, and HR automation, much work on fairness remains to be done. Two complications arise: first, the protected attribute may not be available in many applications. Second, there are multiple measures of fairness of rankings, and optimization-based methods utilizing a single measure of fairness of rankings may produce rankings that are unfair with respect to other measures. In this work, we propose a randomized method for post-processing rankings, which do not require the availability of the protected attribute. In an extensive numerical study, we show the robustness of our methods with respect to P-Fairness and effectiveness with respect to Normalized Discounted Cumulative Gain (NDCG) from the baseline ranking, improving on previously proposed methods.