Research map / Learning theory and algorithms / Combinatorial and search optimization
Tighter Truncated Rectangular Prism Approximation for RNN Robustness Verification
Xingqi Lin, Liangyu Chen, Min Wu, Min Zhang, Zhenbing Zeng
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| replace distance-based LP planes with volume-area hybrid objective | Prover · MNIST (eps=0.012, f=4, h=32, layers=1) | verified accuracy (%): 45 -> 83 | Improved |
| replace distance-based LP planes with volume-area hybrid objective | Prover · MNIST (eps=0.012, f=4, h=32, layers=3) | verified accuracy (%): 25 -> 59 | Improved |
| replace distance-based LP planes with volume-area hybrid objective | Prover · MNIST (eps=0.012, f=4, h=128, layers=1) | verified accuracy (%): 86 -> 93 | Improved |
| volume-area objective cost in single-plane verification | Prover · MNIST (eps=0.012, f=4, h=32, layers=1) | running time (s): 1.47 -> 1.79 | Hurt |
| volume-area objective on speech datasets | Prover · GSC (eps=-100 dB) | verified accuracy (%): 44 -> 46 | Improved |
| volume-area objective on sentiment analysis | Prover · RT sentiment (eps=0.05) | verified accuracy (%): 77 -> 84 | Improved |
| volume-area objective on sentiment analysis, high perturbation | Prover · RT sentiment (eps=0.09) | verified accuracy (%): 27 -> 28 | No clear effect |
| use 4-rectangle division instead of 2-triangle (up) in multi-plane (ablation) | DeepPrism · MNIST multi-plane (eps=0.020) | verified accuracy (%): 18 -> 54 | Improved |
| use finer 16-rectangle division instead of 4-rectangle (ablation) | DeepPrism · MNIST multi-plane (eps=0.020) | verified accuracy (%): 54 -> 64 | Improved |
| multi-plane refinement vs single-plane (ablation) | DeepPrism · MNIST (f=4, h=32, layers=3) | verified accuracy (%) | Improved |
| use different LP approximation objective inside multi-plane refinement (ablation) | multi-plane approximation · MNIST (f=4, h=32, layers=3) | verified accuracy (%) | No clear effect |
Closest papers by meaning
- A Convex Relaxation Barrier to Tight Robustness Verification of Neural Networks (NeurIPS 2019 · similarity 0.87)
- Tightening Robustness Verification of MaxPool-based Neural Networks via Minimizing the Over-Approximation Zone (CVPR 2025 · similarity 0.86)
- Towards General Robustness Verification of MaxPool-based Convolutional Neural Networks via Tightening Linear Approximation (CVPR 2024 · similarity 0.85)
- LEVIS: Large Exact Verifiable Input Spaces for Neural Networks (ICML 2025 · similarity 0.84)
- Interpreting Robustness Proofs of Deep Neural Networks (ICLR 2024 · similarity 0.84)
- VNN: Verification-Friendly Neural Networks with Hard Robustness Guarantees (ICML 2024 · similarity 0.84)
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.
