Generative Gibbs Sampling
A sampling framework that combines multiple pretrained diffusion models with physics-based contextual energy terms. It achieves joint distribution sampling through a Gibbs alternating update strategy, enabling the handling of composite systems without requiring retraining of the constituent models.
Timeline (last 90 days)
The core idea of the generative Gibbs sampling scheme is to combine existing monomer pre-trained models with physical context energy terms describing solvent and ion interactions at the inference stage, rather than training an end-to-end model that covers everything
Proposition 1 proves that if the system is decomposable and each model is perfectly trained on isolated components with the context factor responsible for the remaining energy terms, the joint distribution converges to the true physical distribution in the zero-time limit
All Facts (2)
The core idea of the generative Gibbs sampling scheme is to combine existing monomer pre-trained models with physical context energy terms describing solvent and ion interactions at the inference stage, rather than training an end-to-end model that covers everything
50%UnverifiedProposition 1 proves that if the system is decomposable and each model is perfectly trained on isolated components with the context factor responsible for the remaining energy terms, the joint distribution converges to the true physical distribution in the zero-time limit
50%