[26] Kevin Ellis - Algorithms for Learning to Induce Programs
The Thesis Review
00:00
Innovations in Program Induction with Dreamcoder
This chapter focuses on the Dreamcoder project, detailing its wake-sleep algorithm and the evolution from earlier custom systems. It explores scalable combinatorial searches, highlighting the use of neural networks to generate models and interpret programming tasks, while emphasizing the importance of reusable abstractions. The discussion also touches on the challenges of integrating traditional programming with modern deep learning techniques, reflecting on the potential for improved synthesis methods in complex coding tasks.
Transcript
Play full episode
