Modular Probabilistic Models via Algebraic Effects
summary
Claims a Haskell-embedded DSL that makes probabilistic models first-class and reusable by using algebraic effects so one model serves for simulation, inference, and mixed uses. Implements multimodal models with observable variables consumed from a typed model environment, and expresses simulation/inference as composable program transformations via effect handlers. Demonstrates by composing transition and observation submodels into an HMM and an SIR epidemic model, runs simulation and likelihood-weighting inference, and provides artifacts (evaluated and reusable) that reproduce the results.
AI-written summary · about
chapters (6)
- 0:00Introduction to models
- 1:00Probabilistic programming
- 3:02Introducing propfx
- 4:54Executing models
- 6:40Compositional modeling
- 9:18SIR model application
related talks
26mEffect Handlers for Programmable Inference
17mMachine Learning Primitives as Algebraic Effects
28mA Series of Unfortunate Effects
46mAlgebraic Thinking for evolution of pure functional domain models
28mModular Models of Monoids with Operations
50mFunctional Programming with Effects
01h 06mThe Evolution of Effects
44mEffects, Algebraically yours ..