A simple Sequential Monte Carlo algorithm for posterior approximation

August 6, 2021

Introduction # Sequential Monte Carlo (SMC) can be used to take samples from posterior distributions, as an alternative to popular methods such as Markov Chain Monte Carlo (MCMC) like Metropolis-Hastings, Gibbs Sampling or more state-of-the-art Hamiltonian Monte Carlo (HMC) and No-U-Turn Sampler (NUTS). Instead of creating a single Markov Chain, SMC works by keeping track of a large population of samples, which, in a similar spirit to Genetic Algorithms can be “selected” and “mutated” to evolve a better “fit” to the posterior distribution. ...