Non-Negative Bregman Divergence Minimization for Deep Direct Density Ratio Estimation

TL;DR

This paper introduces a non-negative correction method for Bregman divergence minimization in deep density ratio estimation, reducing overfitting and improving outlier detection performance.

Contribution

It proposes a novel non-negative correction for empirical Bregman divergence estimators to mitigate overfitting in deep density ratio estimation models.

Findings

The method reduces overfitting caused by train-loss hacking.

Theoretical generalization error bounds support the approach.

Experimental results show improved outlier detection performance.

Abstract

Density ratio estimation (DRE) is at the core of various machine learning tasks such as anomaly detection and domain adaptation. In existing studies on DRE, methods based on Bregman divergence (BD) minimization have been extensively studied. However, BD minimization when applied with highly flexible models, such as deep neural networks, tends to suffer from what we call train-loss hacking, which is a source of overfitting caused by a typical characteristic of empirical BD estimators. In this paper, to mitigate train-loss hacking, we propose a non-negative correction for empirical BD estimators. Theoretically, we confirm the soundness of the proposed method through a generalization error bound. Through our experiments, the proposed methods show a favorable performance in inlier-based outlier detection.