3 Hits in 3.6 sec

Beta-CROWN: Efficient Bound Propagation with Per-neuron Split Constraints for Complete and Incomplete Neural Network Robustness Verification [article]

Shiqi Wang, Huan Zhang, Kaidi Xu, Xue Lin, Suman Jana, Cho-Jui Hsieh, J. Zico Kolter
2021 arXiv   pre-print
Bound propagation based incomplete neural network verifiers such as CROWN are very efficient and can significantly accelerate branch-and-bound (BaB) based complete verification of neural networks.  ...  However, bound propagation cannot fully handle the neuron split constraints introduced by BaB commonly handled by expensive linear programming (LP) solvers, leading to loose bounds and hurting verification  ...  Our algorithm empowered the tool α,β-CROWN (alpha-beta-CROWN), which won the 2nd International Verification of Neural Networks Competition [3] (VNN-COMP 2021) with the highest total score and verified  ... 
arXiv:2103.06624v2 fatcat:ier7kpgjnbgpreabp26ik43eda

The Second International Verification of Neural Networks Competition (VNN-COMP 2021): Summary and Results [article]

Stanley Bak, Changliu Liu, Taylor Johnson
2021 arXiv   pre-print
Along this line, we used standard formats (ONNX for neural networks and VNNLIB for specifications), standard hardware (all tools are run by the organizers on AWS), and tool parameters provided by the tool  ...  The goal of the competition is to provide an objective comparison of the state-of-the-art methods in neural network verification, in terms of scalability and speed.  ...  Tool authors listed in section 3 participated in the preparation and review of this report.  ... 
arXiv:2109.00498v1 fatcat:sw2o63xlpndmtlrirfovhtjsx4

Certified Defenses: Why Tighter Relaxations May Hurt Training [article]

Nikola Jovanović, Mislav Balunović, Maximilian Baader, Martin Vechev
2021 arXiv   pre-print
Further, we investigate the possibility of designing and training with relaxations that are tight, continuous and not sensitive.  ...  Certified defenses based on convex relaxations are an established technique for training provably robust models.  ...  Machine Learning and Systems (MLSys), 2021. [24] Beta-crown: Efficient bound propagation with per-neuron split constraints for complete and incomplete neural network verification.  ... 
arXiv:2102.06700v2 fatcat:f3rbasvxc5hdzjhbaolrbpmoea