Understanding Deep Generative Models with Generalized Empirical Likelihoods
This provides a more interpretable evaluation framework for researchers and practitioners working with DGMs, though it is incremental as it builds on existing statistical methods.
The paper tackles the challenge of evaluating deep generative models (DGMs) like GANs and diffusion models, which lack exact likelihoods, by proposing generalized empirical likelihood (GEL) methods as diagnostic tools to identify deficiencies such as mode dropping and imbalance, showing up to 60% better prediction of these issues compared to existing metrics.
Understanding how well a deep generative model captures a distribution of high-dimensional data remains an important open challenge. It is especially difficult for certain model classes, such as Generative Adversarial Networks and Diffusion Models, whose models do not admit exact likelihoods. In this work, we demonstrate that generalized empirical likelihood (GEL) methods offer a family of diagnostic tools that can identify many deficiencies of deep generative models (DGMs). We show, with appropriate specification of moment conditions, that the proposed method can identify which modes have been dropped, the degree to which DGMs are mode imbalanced, and whether DGMs sufficiently capture intra-class diversity. We show how to combine techniques from Maximum Mean Discrepancy and Generalized Empirical Likelihood to create not only distribution tests that retain per-sample interpretability, but also metrics that include label information. We find that such tests predict the degree of mode dropping and mode imbalance up to 60% better than metrics such as improved precision/recall. We provide an implementation at https://github.com/deepmind/understanding_deep_generative_models_with_generalized_empirical_likelihood/.