Abstract:Probabilistic programs with dynamic computation graphs can define measures over sample spaces with unbounded dimensionality, which constitute programmatic analogues to Bayesian nonparametrics. Owing to the generality of this model class, inference relies on `black-box' Monte Carlo methods that are often not able to take advantage of conditional independence and exchangeability, which have historically been the cornerstones of efficient inference. We here seek to develop a `middle ground' between probabilistic models with fully dynamic and fully static computation graphs. To this end, we introduce a combinator library for the Probabilistic Torch framework. Combinators are functions that accept models and return transformed models. We assume that models are dynamic, but that model composition is static, in the sense that combinator application takes place prior to evaluating the model on data. Combinators provide primitives for both model and inference composition. Model combinators take the form of classic functional programming constructs such as map and reduce. These constructs define a computation graph at a coarsened level of representation, in which nodes correspond to models, rather than individual variables. Inference combinators implement operations such as importance resampling and application of a transition kernel, which alter the evaluation strategy for a model whilst preserving proper weighting. Owing to this property, models defined using combinators can be trained using stochastic methods that optimize either variational or wake-sleep style objectives. As a validation of this principle, we use combinators to implement black box inference for hidden Markov models.

Compositional Stochastic Modeling and Probabilistic Programming

Compositional reasoning for probabilistic finite-state behaviors

Composing Modeling and Inference Operations with Probabilistic Program Combinators

Stochastic Probabilistic Programs

Stochastic Constraint Programming

Stochastic Constraint Programming: A Scenario-Based Approach

Compositional Abstractions of Interconnected Discrete-Time Stochastic Control Systems

Integrating Probability with Time and Shared-Variable Concurrency.

PTSC: Probability, Time and Shared-Variable Concurrency

Automatic Generation of Probabilistic Programming from Time Series Data

Linking Theories in Probabilistic Programming

Compositional Inference Metaprogramming with Convergence Guarantees

Discrete-Continuous Mixtures in Probabilistic Programming: Generalized Semantics and Inference Algorithms.

Combining Data and Physical Models for Probabilistic Analysis: A Bayesian Augmented Space Learning Perspective

Compositional Expected Cost Analysis of Functional Probabilistic Programs

Learning and Compositionality: a Unification Attempt via Connectionist Probabilistic Programming

Modular Probabilistic Models via Algebraic Effects

Extending Hybrid CSP with Probability and Stochasticity

Compositional Reasoning for Weighted Markov Decision Processes

Transforming Probabilistic Programs for Model Checking

Compositional imprecise probability