Improving Oracle-Guided Inductive Synthesis by Efficient Question Selection
summary
Proposes LearnSy, a question selector that models complex operator behavior as simple random events and computes the expected number of program pairs that produce the same output to rank candidate counterexamples. Integrates with a verifier by generating candidate inputs, training per-input operator-probability parameters, and picking the input with minimum expected confusion. Evaluated on interactive and non-interactive OGIS/CEGIS tasks, it achieves competitive selection speed and reduces CEGIS solver time by up to 43.0%.
AI-written summary · about
chapters (11)
- 0:00Introduction
- 0:44Program synthesis background
- 2:34Example selection importance
- 4:02Mean pair strategy
- 6:05Overhead in non-interactive settings
- 7:10Efficient grammar calculation
- 9:40Operator behavior approximation
- 11:26Approach integration
- 11:47Evaluation across domains
- 14:15Summary of results
- 14:59Q&A
related talks
15mToward Neural-Network-Guided Program Synthesis and Verification
24mReward-Guided Synthesis of Intelligent Agents with Control Structures
01h 04mInvited Talk: Finding Good Programs by Avoiding Bad Ones
14mHow we taught agents to use good retrieval
23mAn Interactive Learning Environment for Program Design
28mScooPy: Enhancing Program Synthesis with Nested Example Specifications
26mInteractive Theorem Provers for Proof Education
01h 03mImproved Assistance for Interactive Proof