Research map / Adversarial robustness and security / Certified and empirical robustness
FocusBranch: Combinatorial Branch-and-Bound for $\ell_0$ Neural Network Robustness Verification
Renwei Deng, Yang He, Linyi Li, Yuepeng Wang
Read the paperCheck your idea against 137,217 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes FocusBranch, a combinatorial branch-and-bound verifier for l0 local robustness of neural networks that decomposes the l0-ball into focus-pixel-set nodes.
Problem
Verifying local robustness of neural networks under l0 perturbations is combinatorial because a verifier must reason over all sparse subsets of perturbed pixels.
Approach
Decomposes the l0-ball into branch-and-bound nodes, each consisting of a focus pixel set S and a complementary component, so verification proceeds by combinatorially branching over pixel subsets rather than treating the whole l0-ball at once.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add concrete-node branching (adaptive starting level below eps) on top of super-node branching (ablation) | FocusBranch · MNIST afc5, eps=5 | verified instances (of 70): 56 -> 61 | Improved |
| add super-node (covering design) branching on top of concrete-node branching (ablation) | FocusBranch · MNIST afc5, eps=5 | verified instances (of 70): 49 -> 61 | Improved |
| use node abstraction with sampled starting level instead of always starting at level eps (CVD-TT) | FocusBranch vs CVD-TT · l0 verification, eps=5 | verified instances (of 740): 375 -> 641 | Improved |
| increase perturbation budget eps from 3 to 6 | FocusBranch · l0 verification, all networks | verified rate: 99.8% -> 98.8% -> 86.6% (eps 3,4,5) | Hurt |
| batch node bounding in FocusBranch (batching enabled) | FocusBranch on CNNs · l0 verification, eps=5, CNNs | verified instances | Improved |
Closest papers by meaning
- Tightening Robustness Verification of MaxPool-based Neural Networks via Minimizing the Over-Approximation Zone (CVPR 2025 · similarity 0.88)
- Towards General Robustness Verification of MaxPool-based Convolutional Neural Networks via Tightening Linear Approximation (CVPR 2024 · similarity 0.87)
- Certified but Private: Scalable Zero-Knowledge Proofs for the Formal Verification of Neural Networks (NeurIPS 2026 · similarity 0.86)
- Probably Approximately Global Robustness Certification (ICML 2025 · similarity 0.86)
- LEVIS: Large Exact Verifiable Input Spaces for Neural Networks (ICML 2025 · similarity 0.86)
- Clip-and-Verify: Linear Constraint-Driven Domain Clipping for Accelerating Neural Network Verification (NeurIPS 2025 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
