Research map / Adversarial robustness and security / Attacks on text and graphs
Improving Robustness Against Stealthy Weight Bit-Flip Attacks by Output Code Matching
Ozan Özdenizci, Robert Legenstein
Check 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 one-hot outputs with partially overlapping Hadamard output codes | ResNet-20 on CIFAR-10 · CIFAR-10, stealthy T-BFA | bit-flips: 27.91 (8.70) -> 281.75 (115.6) | Improved |
| Replace one-hot outputs with partially overlapping Hadamard output codes | ResNet-50 on ImageNet · ImageNet, stealthy T-BFA | PA-ACC: 68.64 (9.25) -> 50.63 (11.3) | Hurt |
| Increase code length from 16 to 64 (ablation) | OCM ResNet-20 · CIFAR-10, stealthy TA-LBF | ASR: 97.60 -> 72.40 | Improved |
| Increase code length from 16 to 64 (ablation) | OCM ResNet-20 · CIFAR-10, stealthy T-BFA | bit-flips: 95.65 (32.4) -> 281.75 (115.6) | Improved |
| Use 4-bit rather than 8-bit quantization (ablation) | OCM ResNet-20 · CIFAR-10, stealthy T-BFA | bit-flips: 281.75 (115.6) -> 278.48 (110.9) | No clear effect |
| Replace Hadamard codes with random binary codes (ablation) | OCM ResNet-20 · CIFAR-10, stealthy TA-LBF | bit-flips: 73.6 -> 69.9 | Hurt |
| Replace Hadamard codes with random binary codes (ablation) | OCM ResNet-50 · ImageNet, stealthy T-BFA | bit-flips: 121.2 -> 91.1 | Hurt |
| Add Piecewise Clustering to OCM16 (ablation) | ResNet-20 · CIFAR-10 (8-bit), stealthy T-BFA | bit-flips: +17.59 | Improved |
| Add Piecewise Clustering to OCM16 (ablation) | ResNet-20 · CIFAR-10 (8-bit), stealthy T-BFA | PA-ACC: -5.34 | Hurt |
| Add Piecewise Clustering to OCM16 (ablation) | ResNet-20 · CIFAR-10 (4-bit), stealthy TA-LBF | bit-flips: +10.04 | Improved |
| Increase model width from WRN-28-4 to WRN-28-8 (ablation) | Vanilla WRN · CIFAR-100, stealthy T-BFA | bit-flips: 10.11 (4.60) -> 16.14 (4.89) | Improved |
Closest papers by meaning
- Targeted Attack against Deep Neural Networks via Flipping Limited Weight Bits (ICLR 2021 · similarity 0.89)
- Versatile Weight Attack via Flipping Limited Bits (TPAMI 2023 · similarity 0.89)
- Crossfire: An Elastic Defense Framework for Graph Neural Networks Under Bit Flip Attacks (AAAI 2025 · similarity 0.85)
- SilentStriker: Toward Stealthy Bit-Flip Attacks on Large Language Models (NeurIPS 2025 · similarity 0.85)
- Train to Defend: First Defense Against Cryptanalytic Neural Network Parameter Extraction Attacks (NeurIPS 2025 · similarity 0.84)
- Verifying Properties of Binary Neural Networks Using Sparse Polynomial Optimization (ICLR 2025 · 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.
