Research map / Adversarial robustness and security / Certified and empirical robustness
Resilience of Entropy Model in Distributed Neural Networks
Milin Zhang, Mohammad Abdi, Shahriar Rifat, Francesco Restuccia
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 |
|---|---|---|---|
| apply object-aware total variation denoising to PGD-E attacked inputs (eps=16/255) | ResNet-50 with MSHP entropy model · ImageNet PGD-E eps=16/255 | data size (KB): 18.79 -> 13.44 | Improved |
| apply object-aware TV denoising (eps=8/255 and 16/255) | ResNet-50 with MSHP · ImageNet PGD-E | accuracy (%): 52.13 -> 59.87 (eps=16/255); 65.62 -> 67.22 (eps=8/255) | Improved |
| apply object-aware TV denoising on small perturbations (eps=2/255) | ResNet-50 with MSHP · ImageNet PGD-E eps=2/255 | accuracy (%): 73.26 -> 70.82 | Hurt |
| apply defense against adaptive low-frequency and regional attacks | ResNet-50 with MSHP · ImageNet adaptive attacks eps=4/255 | data size (KB): 14.11 -> 9.56 (frequency); 14.05 -> 9.50 (regional) | Improved |
| PGD-E entropy attack versus accuracy attack PGD-Acc | distributed DNN with entropy model · ImageNet eps=4/255 | data size: PGD-E 1.35x bit rate, -2.34% accuracy; PGD-Acc +7.28% size, -42.88% accuracy | Mixed |
| add shot noise corruption (high-frequency) | entropy-coded ResNet-50 · ImageNet-C severity 5 | data size: +65.31% on average | Hurt |
| add defocus blur or contrast corruption | entropy-coded ResNet-50 · ImageNet-C severity 5 | data size: -53.31% and -67.45% on average | Improved |
| use MSHP instead of FP entropy model under PGD-E (eps=8/255) | ResNet-50 · ImageNet eps=8/255 | data size increase under PGD-E: 21.93% (FP) vs 46.82% (MSHP) size increase | Hurt |
| use mask on denoising to restrict to non-object area (ablation) | TV denoising defense · ImageNet | accuracy | Improved |
Closest papers by meaning
- Information Bottleneck Analysis of Deep Neural Networks via Lossy Compression (ICLR 2024 · similarity 0.82)
- Coded Computing for Resilient Distributed Computing: A Learning-Theoretic Framework (NeurIPS 2024 · similarity 0.82)
- Diversify, Contextualize, and Adapt: Efficient Entropy Modeling for Neural Image Codec (NeurIPS 2024 · similarity 0.82)
- Neural Collapse is Globally Optimal in Deep Regularized ResNets and Transformers (NeurIPS 2025 · similarity 0.82)
- Towards a Better Theoretical Understanding of Independent Subnetwork Training (ICML 2024 · similarity 0.82)
- Democratic Training Against Universal Adversarial Perturbations (ICLR 2025 · similarity 0.82)
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.
