Research map / Optimization / Distributed and compressed
Election Coding for Distributed Learning: Protecting SignSGD against Byzantine Attacks
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 Election Coding with redundancy r=2 | SignSGD-MV · CIFAR-10 ResNet-18, n=9, b=3, target test accuracy 79% | training time (minutes): 749.5 -> 351.5 | Helped |
| Add Election Coding with redundancy r=2 | SignSGD-MV · CIFAR-10 ResNet-18, n=9, b=3, target test accuracy 88% | training time (minutes): SignSGD-MV: \infty; Election Coding: 608.7 | Helped |
| Add Election Coding with redundancy r=2.5 | SignSGD-MV · CIFAR-10 ResNet-18, n=5, b=2, target test accuracy 60% | training time (minutes): SignSGD-MV: \infty; Election Coding: 14.0 | Helped |
| Add Election Coding with redundancy r=2.5 | SignSGD-MV · CIFAR-10 ResNet-18, n=5, b=2, target test accuracy 88% | training time (minutes): SignSGD-MV: \infty; Election Coding: 245.0 | Helped |
| Add Election Coding with redundancy r=2.5 | SignSGD-MV · CIFAR-10 ResNet-18, n=5, b=1, target test accuracy 85% | training time (minutes): 191.2 -> 119.0 | Helped |
| Add Election Coding with redundancy r=2.5 | SignSGD-MV · CIFAR-10 ResNet-18, n=5, b=1, target test accuracy 89% | training time (minutes): SignSGD-MV: \infty; Election Coding: 287.0 | Helped |
| Add Election Coding with redundancy r=2 | SignSGD-MV · CIFAR-10 ResNet-18, n=9, b=2, target test accuracy 85% | training time (minutes): 207.5 -> 137.2 | Helped |
| Add Election Coding with redundancy r=2 | SignSGD-MV · CIFAR-10 ResNet-18, n=9, b=2, target test accuracy 89% | training time (minutes): 381.5 -> 334.4 | Helped |
| Reduce mini-batch size to half while using Bernoulli coding (ablation) | Election Coding · CIFAR-10 ResNet-18, n=5 or n=15 | test accuracy: 20\sim30 \% performance gap | Helped |
| Design the deterministic code for fewer Byzantine nodes than are present (ablation) | Election Coding · CIFAR-10 ResNet-18, n=5, actual b=2, designed for \hat{b}=1 | test accuracy: over $20 \%$ | Helped |
| Use Bernoulli Election Coding under the maximum Byzantine setting | SignSGD-MV · CIFAR-10 ResNet-18, n=9, b=4 | test accuracy: over $20\%$ | Helped |
| Add independent Gaussian noise to partition gradients before taking signs (ablation) | Bernoulli Election Coding · CIFAR-10; added noise variance \sigma^2 = 1e^{-8} | test accuracy | Mixed |
Closest papers by meaning
- On the Tension between Byzantine Robustness and No-Attack Accuracy in Distributed Learning (ICML 2025 · similarity 0.89)
- From Inexact Gradients to Byzantine Robustness: Acceleration and Optimization under Similarity (NeurIPS 2026 · similarity 0.88)
- Byzantine Robustness and Partial Participation Can Be Achieved at Once: Just Clip Gradient Differences (NeurIPS 2024 · similarity 0.88)
- Byzantine-Robust Decentralized Learning via Remove-then-Clip Aggregation (AAAI 2024 · similarity 0.87)
- Unified Breakdown Analysis for Byzantine Robust Gossip (ICML 2025 · similarity 0.85)
- RESIST: Resilient Decentralized Learning Using Consensus Gradient Descent (NeurIPS 2026 · similarity 0.85)
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.