A Differentiable Rank-Based Objective For Better Feature Learning

TL;DR

This paper introduces difFOCI, a differentiable approximation of a non-parametric dependence measure, enabling improved feature selection, neural network regularization, and fairness in classification tasks.

Contribution

We develop difFOCI, a differentiable, parametric version of FOCI, allowing broader application in feature learning, neural network training, and fairness without sensitive data.

Findings

difFOCI outperforms FOCI in variable selection tasks

It enhances feature learning and reduces spurious correlations

It can be integrated into neural networks for improved performance

Abstract

In this paper, we leverage existing statistical methods to better understand feature learning from data. We tackle this by modifying the model-free variable selection method, Feature Ordering by Conditional Independence (FOCI), which is introduced in \cite{azadkia2021simple}. While FOCI is based on a non-parametric coefficient of conditional dependence, we introduce its parametric, differentiable approximation. With this approximate coefficient of correlation, we present a new algorithm called difFOCI, which is applicable to a wider range of machine learning problems thanks to its differentiable nature and learnable parameters. We present difFOCI in three contexts: (1) as a variable selection method with baseline comparisons to FOCI, (2) as a trainable model parametrized with a neural network, and (3) as a generic, widely applicable neural network regularizer, one that improves feature learning with better management of spurious correlations. We evaluate difFOCI on increasingly complex problems ranging from basic variable selection in toy examples to saliency map comparisons in convolutional networks. We then show how difFOCI can be incorporated in the context of fairness to facilitate classifications without relying on sensitive data.