Academic

Sampling is one of the most fundamental problems in computational inference. Beyond Bayesian statistics, it has numerous applications in modern inverse problems such as image recovery, protein generation, or probabilistic inference in large language models. In this talk, we argue that state-space models and Sequential Monte Carlo (SMC) methods constitute a natural framework to formulate a wide array of sampling problems, even when no natural temporal component is present. We then describe a variety of strategies to improve the efficiency of SMC, in particular those exploiting recent advances in diffusion-based generative models.