The diffusion Fisher matrix of a Gaussian-perturbed distribution is expressed in the span of data outer products, enabling two faster approximation algorithms for trace and matrix-vector access.
Bounding the Test Log-Likelihood of Generative Models
1 Pith paper cite this work. Polarity classification is still indexing.
abstract
Several interesting generative learning algorithms involve a complex probability distribution over many random variables, involving intractable normalization constants or latent variable normalization. Some of them may even not have an analytic expression for the unnormalized probability function and no tractable approximation. This makes it difficult to estimate the quality of these models, once they have been trained, or to monitor their quality (e.g. for early stopping) while training. A previously proposed method is based on constructing a non-parametric density estimator of the model's probability function from samples generated by the model. We revisit this idea, propose a more efficient estimator, and prove that it provides a lower bound on the true test log-likelihood, and an unbiased estimator as the number of generated samples goes to infinity, although one that incorporates the effect of poor mixing. We further propose a biased variant of the estimator that can be used reliably with a finite number of samples for the purpose of model comparison.
fields
cs.LG 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Efficiently Access Diffusion Fisher: Within the Outer Product Span Space
The diffusion Fisher matrix of a Gaussian-perturbed distribution is expressed in the span of data outer products, enabling two faster approximation algorithms for trace and matrix-vector access.