Conditionally Strongly Log-Concave Generative Models
This addresses the problem of applying generative models to scientific data by bridging theoretical guarantees with practical performance, though it is incremental as it builds on log-concavity concepts.
The paper tackles the gap between deep generative models with mode collapse issues and classical algorithms requiring restrictive assumptions like log-concavity, by introducing conditionally strongly log-concave (CSLC) models that factorize data distributions for efficient estimation and sampling with theoretical guarantees, achieving higher resolution in physical fields like the φ⁴ model and weak lensing maps.
There is a growing gap between the impressive results of deep image generative models and classical algorithms that offer theoretical guarantees. The former suffer from mode collapse or memorization issues, limiting their application to scientific data. The latter require restrictive assumptions such as log-concavity to escape the curse of dimensionality. We partially bridge this gap by introducing conditionally strongly log-concave (CSLC) models, which factorize the data distribution into a product of conditional probability distributions that are strongly log-concave. This factorization is obtained with orthogonal projectors adapted to the data distribution. It leads to efficient parameter estimation and sampling algorithms, with theoretical guarantees, although the data distribution is not globally log-concave. We show that several challenging multiscale processes are conditionally log-concave using wavelet packet orthogonal projectors. Numerical results are shown for physical fields such as the $\varphi^4$ model and weak lensing convergence maps with higher resolution than in previous works.