Explore advanced techniques in neurosymbolic program architecture search in this 35-minute conference talk by Yisong Yue, Swarat Chaudhuri, and Jennifer Sun. Delve into two key methods: Admissible Neural Heuristics (NEAR) for informed graph search and DreamCoder for library learning. Gain insights into practical applications of neurosymbolic learning in behavioral neuroscience. Learn about neural architectures for sequence prediction, differentiable symbolic execution, and safe learning techniques. Understand the integration of verification and learning in neurosymbolic programming. Conclude with an overview of hands-on activities and potential areas for further exploration in this cutting-edge field.
Neurosymbolic Program Architecture Search Methods - Session 3
Neurosymbolic Programming for Science via YouTube
Overview
Syllabus
Outline of Tutorial
Session 3
Recall: Searching over program structures
Basic Idea
Simplest Case: Deterministic Greedy
Next Step: Beam Search
Learning Setup
Neural Architectures for Sequence Prediction
Learning to Search vs. Neural Relaxations
Library Learning
Dreamcoder
Neurosymbolic Programming
Integrate verification and learning
Safe Learning
Verification Technique: Symbolic Execution
DSE: Differentiable Symbolic Execution
Neurosymbolic learning isn't new...
Hands-on Activity Overview
Code Structure
Potential Areas to Explore
Taught by
Neurosymbolic Programming for Science
