Research map / Learning theory and algorithms / Prediction and decision losses
DETOX: A Redundancy-based Framework for Faster and More Robust Gradient Aggregation
Shashank Rajput, Hongyi Wang, Zachary Charles, Dimitris Papailiopoulos
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 DETOX filtering and hierarchical aggregation to Multi-Krum | Multi-Krum · CIFAR-10, ResNet-18, ALIE | test accuracy: 45.24% -> 80.3% | Improved |
| Add DETOX filtering and hierarchical aggregation to Bulyan | Bulyan · CIFAR-100, VGG13-BN, ALIE | test accuracy: 11.06% -> 46.82% | Improved |
| Add DETOX filtering and hierarchical aggregation to coordinate-wise median | Coordinate-wise median · CIFAR-100, VGG13-BN, ALIE | test accuracy: 8.64% -> 59.51% | Improved |
| Pair signSGD with DETOX's redundancy-based filtering | signSGD · CIFAR-10, ResNet-18, constant Byzantine attack | test accuracy: 34.92% -> 78.75% | Improved |
| Pair signSGD with DETOX's redundancy-based filtering | signSGD · CIFAR-100, VGG13-BN, constant Byzantine attack | test accuracy (TOP-1): 2.12% -> 40.37% | Improved |
| Add DETOX to coordinate-wise median | Coordinate-wise median · CIFAR-10, ResNet-18, reverse gradient attack | speedup to 90% test accuracy: 5× | Improved |
| Use DETOX with robust aggregation | Robust aggregators · Distributed training, CIFAR-10 and CIFAR-100 experiments | per-iteration speedup: up to 3× | Improved |
| Use DETOX filtering and hierarchy with coordinate-wise median | Coordinate-wise median · CIFAR-10, ResNet-18, reverse gradient attack | test accuracy | No clear effect |
| Replace mean-at-first-level plus robust final aggregation with Multi-Krum at the first level and mean at the final level (ablation) | DETOX hierarchical aggregation · Byzantine-resilience experiments | test accuracy | Improved |
| Replace mean-at-first-level plus robust final aggregation with Bulyan at the first level and mean at the final level (ablation) | DETOX hierarchical aggregation · Byzantine-resilience experiments | test accuracy | Improved |
| Add redundancy-based majority filtering before robust aggregation | Robust gradient aggregation · Synthetic robust mean estimation | estimation error | Improved |
Closest papers by meaning
- Fault Tolerant ML: Efficient Meta-Aggregation and Synchronous Training (ICML 2024 · similarity 0.92)
- From Inexact Gradients to Byzantine Robustness: Acceleration and Optimization under Similarity (NeurIPS 2026 · similarity 0.92)
- Weight for Robustness: A Comprehensive Approach towards Optimal Fault-Tolerant Asynchronous ML (NeurIPS 2024 · similarity 0.91)
- On the Tension between Byzantine Robustness and No-Attack Accuracy in Distributed Learning (ICML 2025 · similarity 0.91)
- Byzantine-Robust Decentralized Learning via Remove-then-Clip Aggregation (AAAI 2024 · similarity 0.90)
- Flag Aggregator: Scalable Distributed Training under Failures and Augmented Losses using Convex Optimization (ICLR 2024 · similarity 0.89)
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.
