Research map / Optimization / Distributed and compressed
Variance Reduced Median-of-Means Estimator for Byzantine-Robust Distributed Inference
Jiyuan Tu, Weidong Liu, Xiaojun Mao, Xi Chen
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 |
|---|---|---|---|
| replace median with variance reduced median-of-means (VRMOM) aggregation | mean estimator · Gaussian mean estimation p=30, no Byzantine machines | RMSE: 0.0211 -> 0.0175 | Improved |
| replace median with VRMOM aggregation | mean estimator · Gaussian mean estimation p=1, no Byzantine machines | RMSE: 0.0030 -> 0.0025 | Improved |
| replace median with VRMOM aggregation under 15% Byzantine machines | mean estimator · Gaussian mean estimation p=30, Byzantine fraction 0.15 | RMSE: 0.0249 -> 0.0227 | Improved |
| increase number of quantile levels K from 10 to 100 (ablation) | VRMOM · Gaussian mean estimation p=30, no Byzantine machines | RMSE: 0.0175 -> 0.0174 | No clear effect |
| use VRMOM as gradient aggregator in robust CSL instead of MOM | RCSL · Linear regression p=30, no attack | RMSE: 0.0319 -> 0.0231 | Improved |
| use VRMOM as gradient aggregator under Gaussian attack alpha=0.15 | RCSL · Linear regression p=30, Gaussian attack | RMSE: 0.0398 -> 0.0351 | Improved |
| use VRMOM as gradient aggregator under omniscient attack alpha=0.15 | RCSL · Linear regression p=30, omniscient attack | RMSE: 0.0449 -> 0.0396 | Improved |
| use VRMOM as gradient aggregator under bit-flip attack alpha=0.15 | RCSL · Linear regression p=30, bit-flip attack | RMSE: 0.0343 -> 0.0250 | Improved |
| use VRMOM as gradient aggregator in balanced logistic regression | RCSL · Logistic regression p=30, mu_x=0, no Byzantine | RMSE: 0.0699 -> 0.0504 | Improved |
| use VRMOM as gradient aggregator in imbalanced logistic regression | RCSL · Logistic regression p=30, mu_x=0.5, no Byzantine | RMSE: 0.0830 -> 0.0583 | Improved |
| fix iteration count at T=5 instead of T=10 (ablation) | RCSL · Linear regression p=30, no attack | RMSE: 0.0231 -> 0.0231 | No clear effect |
| fix iteration count at T=5 instead of T=10 (ablation) | RCSL · Logistic regression p=30, mu_x=0, alpha=0.15 | RMSE: 0.0702 -> 0.0701 | No clear effect |
Closest papers by meaning
- On the Tension between Byzantine Robustness and No-Attack Accuracy in Distributed Learning (ICML 2025 · similarity 0.87)
- Weight for Robustness: A Comprehensive Approach towards Optimal Fault-Tolerant Asynchronous ML (NeurIPS 2024 · similarity 0.86)
- Byzantine Robustness and Partial Participation Can Be Achieved at Once: Just Clip Gradient Differences (NeurIPS 2024 · similarity 0.86)
- From Inexact Gradients to Byzantine Robustness: Acceleration and Optimization under Similarity (NeurIPS 2026 · similarity 0.86)
- Fault Tolerant ML: Efficient Meta-Aggregation and Synchronous Training (ICML 2024 · similarity 0.85)
- Unified Breakdown Analysis for Byzantine Robust Gossip (ICML 2025 · 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.
