Watch a thought-provoking lecture from Prof. Armando Solar-Lezama exploring how AI will revolutionize programming through self-learning and synthesis. Delve into the transformative potential of AI in code writing and discover how machines can leverage code as a knowledge representation system to become more reliable and introspective. Learn about the synergistic relationship between program synthesis and AI, examining practical applications across diverse domains including cell fate decisions in C. elegans, language morphology and phonology, and controller design for repetitive motions. Explore advanced concepts like syntax-guided synthesis, inductive generalization, and the DreamCoder Algorithm while understanding how program structure and learned subroutines contribute to training performance. Gain insights into the future of software development through demonstrations of LOGO Graphics and discussions on deep hierarchical learning in this 36-minute technical presentation from the INSAIT Tech Series.
AI Will Program Itself: Synthesis, Learning and Beyond
Overview
Syllabus
Intro
Beyond Software Engineering
Syntax Guided Synthesis
Synthesis today
Synthesis but backwards
Modeling systems with synthesis
Cell fate decision in C. elegans
Data: Mutation Experiments
Model synthesis
Language Morphology
Language Phonology
Infering Morphophonology rules
Controllers for repetitive motions
Inductive generalization
Solution: Add program structure
Training performance
Beyond software development
Basic program synthesis
Example: LOGO Graphics
DreamCoder Algorithm
Abstractions in Synthesis
Abstraction Dream
Learned subroutines
Learning deep hierarchies
Domains
Conclusion
Taught by
INSAIT Institute
