Nearest-Neighbor Density Estimation for Dependency Suppression
This work addresses the problem of removing unwanted data dependencies for applications in fairness, robust learning, and privacy protection, offering an incremental improvement over existing methods.
This paper introduces an encoder-based method to remove unwanted dependencies from data while preserving essential characteristics. It achieves this by explicitly estimating and modifying the data distribution using a variational autoencoder and a novel loss function driven by nearest-neighbor density estimation, outperforming existing unsupervised techniques and rivaling supervised methods in balancing information removal and utility.
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.