MultiSPPL: extending SPPL with multivariate leaf nodes
summary
Claims that extending SPPL’s SPN leaves with a SPINNY interface of five methods (sample, PDF, prop, condition, constrain) guarantees closure under conditioning and thus exact inference for trees whose leaves implement it. Demonstrates implementation for multivariate Gaussian and truncated Gaussian leaves using inverse CDF, automatic differentiation, and polytope-constrained sampling, enabling exact linear regression and Kalman filtering. Notes required DSL/type system extensions and new optimization rules for performance; prototypes validate the approach but engineering work remains.
AI-written summary · about
related talks
14mMulti-Language Probabilistic Programming
35mScalable structure learning and inference for domain-specific probabilistic prog
19mGeneralSparse: Bridging the Gap in SpMM for Pruned Large Language Model Inference..
06mPitfalls of Full Bayesian Inference in Universal Probabilistic Programming
56mSparsifying Generalized Linear Models
41mScalable Collective Inference from Richly Structured Data using Probabilistic Soft Logic (PSL)
20mNonparametric Involutive Markov Chain Monte Carlo
14mSequential Monte Carlo Program Synthesis with Refinement Proposals