Probabilistic Programs Which Make (Common) Sense

on YouTube ↗36m

" summary

Claims that Sigma lets programs be written as functional probabilistic models that run forward (render a 3D scene) and in reverse (infer the most probable 3D scene from a 2D image). Implements this in Julia, combining functional programming with statistical inference, program analysis and constraint solving. Demonstrations include sampling posterior bounds, matrix factorization by constraint solving, and crude scene reconstruction; it works approximately, is expressive but slower and not Turing‑universal.

AI-written summary · about

" chapters (12)

1. 0:00Start
2. 0:29Common Sense
3. 4:43How Should We Figure Out Common Sense
4. 5:45Consistency
5. 11:57Demo
6. 15:19Factorize Matrices into Smaller Matrices
7. 17:53Random Variables
8. 20:45Constraint Solver
9. 28:51Constraints on Plausibility
10. 31:31Deep Neural Networks
11. 33:08Convolutional Neural Networks
12. 35:43Representation

" related talks

• 43mThe Power of Probabilistic ProgrammingBen Vigoda · O'Reilly Artificial Intelligence Conference 2017 - New York, New York
• 40mProbabilistic Programming for Augmented IntelligenceVikash K. Mansinghka · Uncertainty in Computation 2016
• 01h 03mA Personal Viewpoint on Probabilistic ProgrammingDaniel Roy · Uncertainty in Computation 2016
• 31mProbabilistic Programming: The What, Why and HowMaria I. Gorinova · SPLASH 2020 REBASE
• 20mProbabilistic Programming with Stochastic ProbabilitiesVikash K. Mansinghka, Martin C. Rinard, Alexander K. Lew, Matin Ghavamizadeh · PLDI 2023
• 29mProbabilistic Programming: What It Is and How It WorksNoel Welsh · Typelevel Summit Philadelphia 2016
• 26mSemi-symbolic Inference for Efficient Streaming Probabilistic ProgrammingCharles Yuan, Michael Carbin, Eric Atkinson, Guillaume Baudart, Louis Mandel · OOPSLA 2022
• 35mScalable structure learning and inference for domain-specific probabilistic progFeras A. Saad · LAFI 2022