2021-07-28 –, Purple
Many Julia libraries implement stochastic simulators of natural and social phenomena, but they are not generally amenable to Bayesian inference. In this talk, we present Genify.jl, which transforms these simulators into the Gen probabilistic programming system via compiler injection, allowing us to compute likelihoods, constrain random variables to specific values, and update these values for Monte Carlo inference, thereby enabling Bayesian inference over a wide range of existing Julia code.
A wide variety of libraries written in Julia implement stochastic simulators of natural and social phenomena for the purposes of computational science. However, these simulators are not generally amenable to Bayesian inference, as they do not provide likelihoods for execution traces, support constraining of observed random variables, or allow random choices and subroutines to be selectively updated in Monte Carlo algorithms.
To address these limitations, we present Genify.jl, an approach to transforming plain Julia code into generative functions in Gen, a universal probabilistic programming system with programmable inference. We accomplish this via lightweight transformation of lowered Julia code into Gen’s dynamic modeling language, combined with a user-friendly random variable addressing scheme that enables straightforward implementation of custom inference programs.
We demonstrate the utility of this approach by transforming an existing agent-based simulator from plain Julia into Gen, and designing custom inference programs that increase accuracy and efficiency relative to generic SMC and MCMC methods. This performance improvement is achieved by proposing, constraining, or re-simulating random variables that are internal to the simulator, which is made possible by transformation into Gen.
Genify.jl is available at: https://github.com/probcomp/Genify.jl
Xuan (Sh-YEN, IPA: ɕɥɛn) is a PhD student at MIT in the Computational Cognitive Science and Probabilistic Computing research groups. Their current research focuses on inferring the hidden structure of human motivations by modeling agents as probabilistic programs, in the hope of aligning AI with the higher-order goals, values, and principles that humans strive (in part) to live by.