Recursion synthesis with unrealizability witnesses
Azadeh Farzan, Danya Lette, Victor Nicolet
摘要
We propose SE2GIS, a novel inductive recursion synthesis approach with the ability to both synthesize code and declare a problem unsolvable. SE2GIS combines a symbolic variant of counterexample-guided inductive synthesis (CEGIS) with a new dual inductive procedure, which focuses on proving a synthesis problem unsolvable rather than finding a solution for it. A vital component of this procedure is a new algorithm that produces a witness, a set of concrete assignments to relevant variables, as a proof that the synthesis instance is not solvable. Witnesses in the dual inductive procedure play the same role that solutions do in classic CEGIS; that is, they ensure progress. Given a reference function, invariants on the input recursive data types, and a target family of recursive functions, SE2GIS synthesizes an implementation in this family that is equivalent to the reference implementation, or declares the problem unsolvable and produces a witness for it. We demonstrate that SE2GIS is effective in both cases; that is, for interesting data types with complex invariants, it can synthesize non-trivial recursive functions or output witnesses that contain useful feedback for the user.
问问这篇 Paper
问问你的智能体。
Lune 读过与它相关的顶会 Paper,每个回答都会注明依据哪几篇。
引用它的顶会 Paper11
- Trace-Guided Inductive Synthesis of Recursive Functional ProgramsYongwei Yuan, Arjun Radhakrishna, Roopsha SamantaPLDI 2023 · 被引用 17 次
- Unrealizability LogicJinwoo Kim, Loris D'Antoni, Thomas W. RepsPOPL 2023 · 被引用 12 次
- Programming-by-Demonstration for Long-Horizon Robot TasksNoah Patton, Kia Rahmani, Meghana Missula, Joydeep Biswas 等POPL 2024 · 被引用 11 次
- Superfusion: Eliminating Intermediate Data Structures via Inductive SynthesisRuyi Ji, Yuwei Zhao, Nadia Polikarpova, Yingfei Xiong 等PLDI 2024 · 被引用 4 次
- Languages with Decidable Learning: A Meta-theoremPaul Krogmeier, P. MadhusudanOOPSLA 2023 · 被引用 4 次
相关 Paper
- Decision Tree Learning in CEGIS-Based Termination AnalysisSatoshi Kura, Hiroshi Unno, Ichiro HasuoCAV 2021 · 被引用 6 次
- Provenance-guided synthesis of Datalog programsMukund Raghothaman, Jonathan Mendelson, David Zhao, Mayur Naik 等POPL 2020 · 被引用 49 次
- Counterexample-Guided Partial Bounding for Recursive Function SynthesisAzadeh Farzan, Victor NicoletCAV 2021 · 被引用 13 次
- Combining the top-down propagation and bottom-up enumeration for inductive program synthesisWoosuk LeePOPL 2021 · 被引用 34 次
- Learning to Synthesize Relational InvariantsJingbo Wang, Chao WangASE 2022 · 被引用 9 次
