Beta-CROWN: Efficient Bound Propagation with Per-neuron Split Constraints for Neural Network Robustness Verification
Dez 6, 2021
Sprecher:innen
Shiqi Wang
Sprecher:in · 0 Follower:innen
Huan Zhang
Sprecher:in · 0 Follower:innen
Suman Jana
Sprecher:in · 0 Follower:innen
Cho-Jui Hsieh
Sprecher:in · 0 Follower:innen
Zico Kolter
Sprecher:in · 2 Follower:innen
Über
Recent works in neural network verification show that cheap incomplete verifiers such as CROWN can be used in Branch-and-Bound (BaB) methods and significantly accelerate complete verification on GPUs. However, they cannot fully handle the neuron split constraints introduced by BaB that are commonly handled by expensive linear programming (LP) solvers, leading to looser bounds and hurting verification efficiency. In this work, we develop β-CROWN, a new bound propagation based method that can fully encode neuron splits via optimizable parameters β constructed from either primal or dual space. When jointly optimized in intermediate layers, β-CROWN generally produces better bounds than typical LP verifiers with neuron split constraints, while being as efficient and parallelizable as CROWN on GPUs. Applied to complete robustness verification benchmarks, β-CROWN with BaB is close to three orders of magnitude faster than LP-based BaB methods, and is at least 3 times faster than winners of VNN-COMP 2020 competition, and also produces lower timeout rates. By terminating BaB early, our method can also be used for incomplete verification. We achieve significantly higher verified accuracy over most existing popular incomplete verifiers and non-trivial improvements over concurrent work PRIMA. Even compared to the tightest but very costly incomplete verifiers SDP-FO, we can obtain higher verified accuracy within over three orders of magnitudes less verification time.Recent works in neural network verification show that cheap incomplete verifiers such as CROWN can be used in Branch-and-Bound (BaB) methods and significantly accelerate complete verification on GPUs. However, they cannot fully handle the neuron split constraints introduced by BaB that are commonly handled by expensive linear programming (LP) solvers, leading to looser bounds and hurting verification efficiency. In this work, we develop β-CROWN, a new bound propagation based method that can full…
Organisator
NeurIPS 2021
Konto · 1,9k Follower:innen
Über NeurIPS 2021
Neural Information Processing Systems (NeurIPS) is a multi-track machine learning and computational neuroscience conference that includes invited talks, demonstrations, symposia and oral and poster presentations of refereed papers. Following the conference, there are workshops which provide a less formal setting.
Gefällt euch das Format? Vertraut auf SlidesLive, um euer nächstes Event festzuhalten!
Professionelle Aufzeichnung und Livestreaming – weltweit.
Freigeben
Empfohlene Videos
Präsentationen, deren Thema, Kategorie oder Sprecher:in ähnlich sind