The Power of Probabilistic Programming
summary
Gamalon presents a concrete pipeline: build a Python simulation of how data is generated, declare uncertain pieces as random variables, then run an automated solver that evolves mixtures of sampling, variational and gradient methods to infer the posterior. The system—born from DARPA work—brands this Bayesian Program Synthesis and calls the interactive layer idea learning. Demonstrated uses include a sketch-to-shape prototype, exoplanet atmosphere inversion, and a product for structures unstructured text into rows or API calls.
AI-written summary · about
chapters (14)
- 0:00Intro
- 1:33Bens background
- 5:43Gamelan
- 10:09Random corners
- 13:11Simulation
- 15:23What is probabilistic programming
- 17:32Monte Carlo vs variational methods
- 19:57No free lunch theorem
- 28:16Design thinking process
- 30:42Simulations
- 36:49Idea Learning
- 39:43Learning More
- 40:21seminal papers
- 40:54outro
related talks
01h 03mAn Overview of Probabilistic Programming
31mProbabilistic Programming: The What, Why and How
40mProbabilistic Programming for Augmented Intelligence
01h 03mA Personal Viewpoint on Probabilistic Programming
32mProbabilistic and Differentiable Programming in Scientific Simulators
01h 07mProbabilistic Programming and Readable Models
47mThe State of the Art for Probabilistic Programming
38mProbabilistic programming and meta-programming in Clojure