Toward neural-network-guided program synthesis and verification

DOI10.1007/978-3-030-88806-0_12MaRDI QIDQ2145332zbMATH OpenOpenAlexFDO

Authors Naoki Kobayashi, Taro Sekiyama, Issei Sato, Hiroshi Unno

Publication date 17 June 2022

Full work available at URL https://arxiv.org/abs/2103.09414

Mathematics Subject Classification ID

Learning and adaptive systems in artificial intelligence (68T05) Mathematical aspects of software engineering (specification, verification, metrics, requirements, etc.) (68N30)

Abstract: We propose a novel framework of program and invariant synthesis called neural network-guided synthesis. We first show that, by suitably designing and training neural networks, we can extract logical formulas over integers from the weights and biases of the trained neural networks. Based on the idea, we have implemented a tool to synthesize formulas from positive/negative examples and implication constraints, and obtained promising experimental results. We also discuss two applications of our synthesis method. One is the use of our tool for qualifier discovery in the framework of ICE-learning-based CHC solving, which can in turn be applied to program verification and inductive invariant synthesis. Another application is to a new program development framework called oracle-based programming, which is a neural-network-guided variation of Solar-Lezama's program synthesis by sketching.

Recommendations

• Code2Inv: a deep learning framework for program verification
• Example Guided Synthesis of Linear Approximations for Neural Network Verification
• From program verification to program synthesis
• Learning abstractions for program synthesis
• Programmable program synthesis

Cites work

• Horn clause solvers for program verification
• ICE-based refinement type discovery for higher-order functional programs
• Learning safe neural network controllers with barrier certificates
• Overfitting in synthesis: theory and practice
• Reluplex: an efficient SMT solver for verifying deep neural networks
• SMT-based model checking for recursive programs
• Safety verification of deep neural networks

Cited in

(6)

• Guiding an automated theorem prover with neural rewriting
• Code2Inv: a deep learning framework for program verification
• scientific article; zbMATH DE number 2080130 (Why is no real title available?)
• Example Guided Synthesis of Linear Approximations for Neural Network Verification
• Challenging SMT solvers to verify neural networks
• Neural network-guided synthesis of recursive list functions

Describes a project that uses

Uses Software

• CLN2INV
• AI2
• nncontroller
• Reluplex
• Adam
• Spacer

This page was built for publication: Toward neural-network-guided program synthesis and verification

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q2145332)