Scholarlink

We present NeuralSAT, a DNN verification tool based on the DPLL(T) framework in SAT solving with conflict clause learning. NeuralSAT participated in VNN-COMP’23 and VNN-COMP’24, with recent improvements such as parallel DPLL(T) and neuron stabilization optimizations. The theoretical foundations and algorithmic details of NeuralSAT are described in prior work, and this paper focuses on the engineering aspects of NeuralSAT, including its design, configuration, and performance in the context of the VNN-COMP evaluation framework. NeuralSAT is available at: https://github.com/dynaroars/neuralsat .

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

NeuralSAT: Scaling Constraint Solving for DNN Verification (Competition Contribution)

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Hai Duong,
  • ThanhVu Nguyen

摘要

We present NeuralSAT, a DNN verification tool based on the DPLL(T) framework in SAT solving with conflict clause learning. NeuralSAT participated in VNN-COMP’23 and VNN-COMP’24, with recent improvements such as parallel DPLL(T) and neuron stabilization optimizations. The theoretical foundations and algorithmic details of NeuralSAT are described in prior work, and this paper focuses on the engineering aspects of NeuralSAT, including its design, configuration, and performance in the context of the VNN-COMP evaluation framework. NeuralSAT is available at: https://github.com/dynaroars/neuralsat .