Effect Handlers for Programmable Inference - Haskell 2023

Explore a 26-minute conference talk from Haskell 2023 that delves into using effect handlers for programmable inference in probabilistic programming. Learn how algebraic effects can provide a structured and modular foundation for inference algorithms, offering an alternative to monad transformers. Discover two abstract algorithms representing Metropolis-Hastings and particle filtering, and see how this approach reveals high-level structure and facilitates easy customization. Gain insights into implementing these inference patterns as a Haskell library and understand the advantages and disadvantages of algebraic effects compared to monad transformers in modular imperative algorithm design.

Introduction
Inferring Missing Data
Algebraic Effects
Framework
Inference
Independence Metropolis
Particle Filters
Multinomial Particle Filter
Resample Move Particle Filter
Recap
Questions