mlsanityData to 14 Oct 2026

Verifying Neural Networks with Reinforcement Learning

arXiv 2609.34553 ↗First posted 28 Sep 2026cs.LG cs.SEcited by 0

On the map: Deep reinforcement learning methods → RL training of language and agent models

Abstract (arXiv)

Formal verification can play a key role in ensuring the reliability of Deep Neural Networks (DNNs) deployed in safety-critical systems. Modern DNN verifiers employ a branch-and-bound framework, which alternates between branching (splitting into smaller subproblems) and bounding (pruning subproblems) to efficiently explore the verification space. However, existing branching heuristics make greedy decisions based on static scoring functions. They do not anticipate long-term efficiency or leverage the growing availability of verification data to improve performance. This work introduces RSB, a reinforcement learning framework that learns to refine baseline branching heuristics. It trains an actor-critic architecture to maximize cumulative future rewards rather than immediate scores. The actor generates attention weights from observations of raw neuron features and learned graph embeddings, which rescale baseline heuristic scores to guide neuron branching. Evaluation on 600 challenging instances demonstrates that RSB consistently outperforms state-of-the-art branching heuristics, solving 11% more instances while reducing branch exploration by 50%.

Key facts. No citing papers in the map yet (data through 14 Oct 2026). Placed in RL training of language and agent models, one of 15 directions in Deep reinforcement learning methods. Builds on 32 of its 51 references that are in the map. First posted on arXiv 28 Sep 2026.

Builds on

32 of 51 references.

References · 32

  1. 2017Proximal Policy Optimization Algorithmscited by 15,215
  2. 2016Semi-Supervised Classification with Graph Convolutional Networkscited by 11,361
  3. 2020Scaling Laws for Neural Language Modelscited by 6,586
  4. 2017Towards Deep Learning Models Resistant to Adversarial Attackscited by 6,415
  5. 2018Soft Actor-Critic: Off-Policy Maximum Entropy Deep Reinforcement Learning with a Stochastic Actorcited by 4,885
  6. 2016Asynchronous Methods for Deep Reinforcement Learningcited by 2,258
  7. 2020Offline Reinforcement Learning: Tutorial, Review, and Perspectives on Open Problemscited by 1,920
  8. 2017Reluplex: An Efficient SMT Solver for Verifying Deep Neural Networkscited by 744
  9. 2016Neural Combinatorial Optimization with Reinforcement Learningcited by 608
  10. 2018Efficient Neural Network Robustness Certification with General Activation Functionscited by 445
  11. 2022Safety-Enhanced Autonomous Driving Using Interpretable Sensor Fusion Transformercited by 212
  12. 2020Fast and Complete: Enabling Complete Neural Network Verification with Rapid and Massively Parallel Incomplete Verifierscited by 187
20 more
  1. 2018Formal Security Analysis of Neural Networks using Symbolic Intervalscited by 185
  2. 2019Branch and Bound for Piecewise Linear Neural Network Verificationcited by 138
  3. 2022General Cutting Planes for Bound-Propagation-Based Neural Network Verificationcited by 103
  4. 2022Complete Verification via Multi-Neuron Relaxation Guided Branch-and-Boundcited by 92
  5. 2024Marabou 2.0: A Versatile Formal Analyzer of Neural Networkscited by 81
  6. 2021The Second International Verification of Neural Networks Competition (VNN-COMP 2021): Summary and Resultscited by 71
  7. 2023First Three Years of the International Verification of Neural Networks Competition (VNN-COMP)cited by 58
  8. 2019Neural Network Branching for Neural Network Verificationcited by 45
  9. 2021Improved Branch and Bound for Neural Network Verification via Lagrangian Decompositioncited by 44
  10. 2024The Fifth International Verification of Neural Networks Competition (VNN-COMP 2024): Summary and Resultscited by 43
  11. 2020Parallelization Techniques for Verifying Neural Networkscited by 36
  12. 2024Scalable Neural Network Verification with Branch-and-bound Inferred Cutting Planescited by 33
  13. 2025The 6th International Verification of Neural Networks Competition (VNN-COMP 2025): Summary and Resultscited by 27
  14. 2023A DPLL(T) Framework for Verifying Deep Neural Networkscited by 24
  15. 2022IBP Regularization for Verified Adversarial Robustness via Branch-and-Boundcited by 21
  16. 2021NeuroBack: Improving CDCL SAT Solving using Graph Neural Networkscited by 19
  17. 2024Harnessing Neuron Stability to Improve DNN Verificationcited by 13
  18. 2021Characterizing the Gap Between Actor-Critic and Policy Gradientcited by 9
  19. 2025Actor-Critics Can Achieve Optimal Sample Efficiencycited by 4
  20. 2025Learning from Algorithm Feedback: One-Shot SAT Solver Guidance with GNNscited by 4

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2022Deep Binary Reinforcement Learning for Scalable Verification
  2. 2024Formally Verifying Deep Reinforcement Learning Controllers with Lyapunov Barrier Certificates
  3. 2025SymDQN: Symbolic Knowledge and Reasoning in Neural Network-based Reinforcement Learning
  4. 2026EvolveGen: Algorithmic Level Hardware Model Checking Benchmark Generation through Reinforcement Learning
  5. 2024Training Verifiably Robust Agents Using Set-Based Reinforcement Learning
  6. 2026A Survey on the Verification of Reinforcement Learning Policies
  7. 2026Probabilistic Verification of Recurrent Neural Networks for Single and Multi-Agent Reinforcement Learning
  8. 2025VerifyBench: Benchmarking Reference-based Reward Systems for Large Language Models