Structured Deep Learning with Probabilistic Neural Prog
summary
Claims an open source Scala library called Probabilistic Neural Programs that combines a probability monad with neural nets to let a model define its output space and scoring at once. Demonstrations include coin flips and a binomial distribution, a sentiment classifier trained on ~11k movie-review sentences with a 100‑dim CBOW encoder-classifier that trains in ~15s, and a natural-language calculator that generates well‑formed prefix expressions via an LSTM scanner feeding per-op vectors, producing executable programs and sensible probabilities.
AI-written summary · about
chapters (8)
- 0:00Introduction
- 1:01Motivation
- 3:15The library
- 4:55Outline
- 9:54Output Space
- 20:27Complex Distributions
- 25:42Assembly
- 29:39Summary
related talks
01h 03mAn Overview of Probabilistic Programming
01h 03mA Personal Viewpoint on Probabilistic Programming
40mProbabilistic Programming for Augmented Intelligence
43mThe Power of Probabilistic Programming
38mProbabilistic programming and meta-programming in Clojure
30mProbabilistic modelling with Scala
37mMonadic Deep Learning
31mProbabilistic Programming: The What, Why and How