MLLGJun 28, 2019

Empirical Study of the Benefits of Overparameterization in Learning Latent Variable Models

arXiv:1907.00030v322 citations
Originality Incremental advance
AI Analysis

This addresses the problem of improving unsupervised learning for researchers and practitioners by showing incremental benefits of overparameterization in latent variable models.

The study empirically investigates the benefits of overparameterization in unsupervised learning of latent variable models, finding that it significantly increases the number of ground truth latent variables recovered across various models and training algorithms.

One of the most surprising and exciting discoveries in supervised learning was the benefit of overparameterization (i.e. training a very large model) to improving the optimization landscape of a problem, with minimal effect on statistical performance (i.e. generalization). In contrast, unsupervised settings have been under-explored, despite the fact that it was observed that overparameterization can be helpful as early as Dasgupta & Schulman (2007). We perform an empirical study of different aspects of overparameterization in unsupervised learning of latent variable models via synthetic and semi-synthetic experiments. We discuss benefits to different metrics of success (recovering the parameters of the ground-truth model, held-out log-likelihood), sensitivity to variations of the training algorithm, and behavior as the amount of overparameterization increases. We find that across a variety of models (noisy-OR networks, sparse coding, probabilistic context-free grammars) and training algorithms (variational inference, alternating minimization, expectation-maximization), overparameterization can significantly increase the number of ground truth latent variables recovered.

Code Implementations1 repo
Foundations

The foundational work for this paper's niche, ranked by how specifically the neighbourhood builds on it — not by global fame.

Your Notes