Sequential Monte Carlo

A class of Bayesian inference methods based on particle filtering that approximate complex distributions by maintaining and iteratively resampling a set of weighted particles. The approach has been applied to LLM inference-time computation to improve sampling quality.