Information-Geometric Decomposition of Generalization Error in Unsupervised Learning

TL;DR

This paper presents an exact information-geometric decomposition of the generalization error in unsupervised learning, applied specifically to epsilon-PCA, revealing optimal model complexity and phase transitions.

Contribution

It introduces an exact decomposition of the KL divergence-based generalization error into interpretable components for e-flat models, with a detailed analysis of epsilon-PCA.

Findings

Decomposition into model error, data bias, and variance components.

Optimal cutoff eigenvalue at epsilon, balancing error and bias.

Identification of phase regimes: retain-all, interior, and collapse.

Abstract

We decompose the Kullback--Leibler generalization error (GE) -- the expected KL divergence from the data distribution to the trained model -- of unsupervised learning into three non-negative components: model error, data bias, and variance. The decomposition is exact for any e-flat model class and follows from two identities of information geometry: the generalized Pythagorean theorem and a dual e-mixture variance identity. As an analytically tractable demonstration, we apply the framework to $\epsilon$-PCA, a regularized principal component analysis in which the empirical covariance is truncated at rank $N_K$ and discarded directions are pinned at a fixed noise floor $\epsilon$. Although rank-constrained $\epsilon$-PCA is not itself e-flat, it admits a technical reformulation with the same total GE on isotropic Gaussian data, under which each component of the decomposition takes closed form. The optimal rank emerges as the cutoff $\lambda_{\mathrm{cut}}^{*} = \epsilon$ -- the model retains exactly those empirical eigenvalues exceeding the noise floor -- with the cutoff reflecting a marginal-rate balance between model-error gain and data-bias cost. A boundary comparison further yields a three-regime phase diagram -- retain-all, interior, and collapse -- separated by the lower Marchenko--Pastur edge and an analytically computable collapse threshold $\epsilon_{*}(\alpha)$, where $\alpha$ is the dimension-to-sample-size ratio. All claims are verified numerically.