Research map / Optimization / Distributed and compressed
Buffered Asynchronous SGD for Byzantine Learning
Yi-Rui Yang, Wu-Jun Li
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 momentum to BASGD (BASGDm instead of BASGD) (ablation) | BASGD · CIFAR-10 ResNet-20 under FoE and ALIE omniscient attacks | test accuracy | Improved |
| increase number of buffers B in BASGD (ablation) | BASGD · CIFAR-10 ResNet-20, no attack | test accuracy | Hurt |
| use BASGD buffers instead of Kardam filtering | asynchronous Byzantine learning · CIFAR-10 ResNet-20, RD-attack and NG-attack | test accuracy | Improved |
| use BASGDm instead of Kardam | asynchronous Byzantine learning · CIFAR-10 ResNet-20, FoE and ALIE attacks | test accuracy | Mixed |
| use BASGD with trmean instead of Kardam | asynchronous Byzantine learning · WikiText-2 LSTM, FoE and ALIE attacks | perplexity | Hurt |
| use BASGDm with any aggregation rule instead of Kardam | asynchronous Byzantine learning · WikiText-2 LSTM, all four attacks | perplexity | Improved |
| use BASGDm (B=10) with trmean instead of Kardam gamma=10 | asynchronous Byzantine learning · CIFAR-10 ResNet-20, no attack | wall-clock time (s): 1535.22 -> 1191.01 | Improved |
| increase Kardam assumed Byzantine count gamma (ablation) | Kardam · NG-attack, 3 Byzantine workers | total filtered loyal gradients: 51.12% -> 56.53% -> 89.07% | Hurt |
Closest papers by meaning
- On the Tension between Byzantine Robustness and No-Attack Accuracy in Distributed Learning (ICML 2025 · similarity 0.88)
- Weight for Robustness: A Comprehensive Approach towards Optimal Fault-Tolerant Asynchronous ML (NeurIPS 2024 · similarity 0.87)
- From Inexact Gradients to Byzantine Robustness: Acceleration and Optimization under Similarity (NeurIPS 2026 · similarity 0.86)
- Byzantine-Robust Decentralized Learning via Remove-then-Clip Aggregation (AAAI 2024 · similarity 0.86)
- Ringmaster ASGD: The First Asynchronous SGD with Optimal Time Complexity (ICML 2025 · similarity 0.85)
- Byzantine Robustness and Partial Participation Can Be Achieved at Once: Just Clip Gradient Differences (NeurIPS 2024 · 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.
