Nearest-Neighbor Density Estimation for Dependency Suppression

TL;DR

This paper introduces a novel encoder-based method utilizing nearest-neighbor density estimation to effectively remove dependencies from data, enhancing fairness, privacy, and robustness without relying on adversarial training.

Contribution

It presents a new approach combining variational autoencoders with non-parametric density estimation to explicitly and directly optimize data independence from sensitive variables.

Findings

Outperforms existing unsupervised dependency removal techniques

Rivals supervised methods in balancing data utility and independence

Effective across multiple datasets and applications

Abstract

The ability to remove unwanted dependencies from data is crucial in various domains, including fairness, robust learning, and privacy protection. In this work, we propose an encoder-based approach that learns a representation independent of a sensitive variable but otherwise preserving essential data characteristics. Unlike existing methods that rely on decorrelation or adversarial learning, our approach explicitly estimates and modifies the data distribution to neutralize statistical dependencies. To achieve this, we combine a specialized variational autoencoder with a novel loss function driven by non-parametric nearest-neighbor density estimation, enabling direct optimization of independence. We evaluate our approach on multiple datasets, demonstrating that it can outperform existing unsupervised techniques and even rival supervised methods in balancing information removal and utility.