Automatic Alignment in Higher-Order Probabilistic Programming Languages

Probabilistic Programming Languages (PPLs) allow users to encode statistical inference problems and automatically apply an algorithm solve them. Popular algorithms for PPLs, such as sequential Monte Carlo (SMC) Markov chain (MCMC), are built around checkpoints -- relevant events the during execution of a probabilistic program. Deciding location is, in current not done optimally. To this problem, we present static analysis technique that determines programs, relieving PPL task. The identifies set execute same order every program run they aligned. We formalize alignment, prove correctness analysis, implement part higher-order functional Miking CorePPL. By utilizing alignment design two novel variants: aligned SMC lightweight MCMC. show, through real-world experiments, significantly improve time accuracy compared standard versions