A New Look on Diffusion Times for Score-based Generative Models

Giulio Franzese (1) , Simone Rossi (1) , Lixuan Yang (2) , Alessandro Finamore (2) , Dario Rossi (2) , Maurizio Filippone (1) , Pietro Michiardi (1)

1 Eurecom [Sophia Antipolis]
2 Huawei Technologies France

Score-based diffusion models map noise into data using stochastic differential equations. While current practice advocates for a large T to ensure closeness to steady state, a smaller value of T should be preferred for a better approximation of the score-matching objective and computational efficiency. We conjecture, contrary to current belief and corroborated by numerical evidence, that the optimal diffusion times are smaller than current practice.

Domaines: Ingénierie assistée par ordinateur

Giulio Franzese, Simone Rossi, Lixuan Yang, Alessandro Finamore, Dario Rossi, et al.. A New Look on Diffusion Times for Score-based Generative Models. ICML 2022, 39th International Conference on Machine Learning, Continuous time methods for machine learning Workshop, Jul 2022, Baltimore, United States. ⟨hal-03889654⟩