Deep Learning Foundations: Xinyun Chen 's talk on "Learning-Based Program Synthesis"

Course webpage: www.cs.umd.edu/...
With the advancement of modern technologies, programming becomes ubiquitous not only among professional software developers, but also for general computer users. However, gaining programming expertise is time-consuming and challenging. Therefore, program synthesis has many applications, where the computer automatically synthesizes programs from specifications such as natural language descriptions and input-output examples. In this talk, I will present my work on learning-based program synthesis, where we have developed deep learning techniques to handle 3 core challenges of program synthesis: input ambiguity, program complexity, and generalization. The talk will have three parts: (1) learning to synthesize programs from multi-modal and potentially ambiguous specifications in the wild; (2) learning with execution for solving challenging programming problems; and (3) compositional generalization via program learning.
Readings:
1. SpreadsheetCoder: Formula Prediction from Semi-structured Context: arxiv.org/pdf/...
2. Execution-Guided Neural Program Synthesis: openreview.net...
3. Compositional Generalization via Neural-Symbolic Stack Machines: arxiv.org/pdf/...
4. (additional) Competition-Level Code Generation with AlphaCode: arxiv.org/pdf/...
5. (additional) Compositional Semantic Parsing with Large Language Models: arxiv.org/pdf/...