A Stable Neural Statistical Dependence Estimator for Autoencoder Feature Analysis

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Hu, Bo, Principe, Jose C
Natura: Preprint
Pubblicazione: 2026
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866912977194582016
author Hu, Bo
Principe, Jose C
author_facet Hu, Bo
Principe, Jose C
contents Statistical dependence measures like mutual information is ideal for analyzing autoencoders, but it can be ill-posed for deterministic, static, noise-free networks. We adopt the variational (Gaussian) formulation that makes dependence among inputs, latents, and reconstructions measurable, and we propose a stable neural dependence estimator based on an orthonormal density-ratio decomposition. Unlike MINE, our method avoids input concatenation and product-of-marginals re-pairing, reducing computational cost and improving stability. We introduce an efficient NMF-like scalar cost and demonstrate empirically that assuming Gaussian noise to form an auxiliary variable enables meaningful dependence measurements and supports quantitative feature analysis, with a sequential convergence of singular values.
format Preprint
id arxiv_https___arxiv_org_abs_2603_11428
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle A Stable Neural Statistical Dependence Estimator for Autoencoder Feature Analysis
Hu, Bo
Principe, Jose C
Machine Learning
Artificial Intelligence
Statistical dependence measures like mutual information is ideal for analyzing autoencoders, but it can be ill-posed for deterministic, static, noise-free networks. We adopt the variational (Gaussian) formulation that makes dependence among inputs, latents, and reconstructions measurable, and we propose a stable neural dependence estimator based on an orthonormal density-ratio decomposition. Unlike MINE, our method avoids input concatenation and product-of-marginals re-pairing, reducing computational cost and improving stability. We introduce an efficient NMF-like scalar cost and demonstrate empirically that assuming Gaussian noise to form an auxiliary variable enables meaningful dependence measurements and supports quantitative feature analysis, with a sequential convergence of singular values.
title A Stable Neural Statistical Dependence Estimator for Autoencoder Feature Analysis
topic Machine Learning
Artificial Intelligence
url https://arxiv.org/abs/2603.11428