A slice sampler for the Indian Buffet Process: expressivity in nonparametric pro...
summary
Introduces a slice-sampling construction that encodes the Indian Buffet Process inside a pure probabilistic program without truncation or hidden state, by adding a slice variable S so the customer conditional is nonzero for only finitely many dishes and becomes uniform on that finite set. Uses MCMC weighting to sample S from an unnormalized infinite-product density and relies on nested inference implemented in Lazy PPL with lazy infinite lists; reports basic experiments that succeed but note nested-inference correctness and efficiency remain challenging.
AI-written summary · about
related talks
29mProbabilistic Programming: What It Is and How It Works
01h 03mA Personal Viewpoint on Probabilistic Programming
14mMulti-Language Probabilistic Programming
20mWhat do posterior distributions of probabilistic programs look like?
01h 02mNonparametric Bayesian Methods: Models, Algorithms, and Applications III
20mProbabilistic Programming with Stochastic Probabilities
01h 04mNonparametric Bayesian Methods: Models, Algorithms, and Applications II
43mPrivacy Amplification by Sampling in Practice