Research map / Adversarial robustness and security / Certified and empirical robustness
Towards Certificated Model Robustness Against Weight Perturbations
Tsui-Wei Weng, Pu Zhao, Sijia Liu, Pin-Yu Chen, Xue Lin, Luca Daniel
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 |
|---|---|---|---|
| add certification constraint (epsilon at a percentile of certified bounds) to ADMM weight quantization (ablation) | 8-layer MLP quantized (4/6/8 bits) · SVHN | test accuracy: around 10% improvement | Improved |
| increase certification epsilon from 100th percentile to 10 times that bound (ablation) | certified quantization · 6-layer MLP 6-bit | test accuracy (%): 83% -> 76.1% | Hurt |
| use proposed ADMM (auxiliary variable on V, includes q) instead of LB method even without certification | ADMM quantization · MNIST-Fashion and SVHN | training accuracy | Improved |
| add certification constraints to the proposed ADMM quantization (ablation) | ADMM quantization · MNIST-Fashion and SVHN | training and test accuracy | Improved |
| choice of epsilon percentile in certification constraint (below 90th percentile) (ablation) | certified quantization · SVHN 8-layer MLP | test accuracy | No clear effect |
Closest papers by meaning
- Formalizing Generalization and Adversarial Robustness of Neural Networks to Weight Perturbations (NeurIPS 2021 · similarity 0.90)
- Theory of Minimal Weight Perturbations in Deep Networks and its Applications for Low-Rank Activated Backdoor Attacks (ICML 2026 · similarity 0.90)
- Quantization-Aware Interval Bound Propagation for Training Certifiably Robust Quantized Neural Networks (AAAI 2023 · similarity 0.88)
- Adversarial Robustness vs. Model Compression, or Both? (ICCV 2019 · similarity 0.88)
- Attacking Binarized Neural Networks (ICLR 2018 · similarity 0.88)
- Learning Better Certified Models from Empirically-Robust Teachers (NeurIPS 2026 · similarity 0.87)
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.
