Generative diffusion posterior sampling for informative likelihoods
This work addresses a specific challenge in sequential Monte Carlo methods for diffusion models, representing an incremental improvement in sampling efficiency.
The paper tackles the problem of conditional sampling with generative diffusion models under outlier conditions or highly informative likelihoods by proposing a new diffusion posterior SMC sampler, achieving improved statistical efficiencies as shown in empirical results.
Sequential Monte Carlo (SMC) methods have recently shown successful results for conditional sampling of generative diffusion models. In this paper we propose a new diffusion posterior SMC sampler achieving improved statistical efficiencies, particularly under outlier conditions or highly informative likelihoods. The key idea is to construct an observation path that correlates with the diffusion model and to design the sampler to leverage this correlation for more efficient sampling. Empirical results conclude the efficiency.