Research map / Optimization / Distributed and compressed
RSA: Byzantine-Robust Stochastic Aggregation Methods for Distributed Learning from Heterogeneous Datasets
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 |
|---|---|---|---|
| use ℓ1-norm RSA instead of Ideal SGD | MNIST softmax regression without Byzantine workers · MNIST test set, no attacks | top-1 accuracy | No clear effect |
| use ℓ1-norm RSA instead of gradient-aggregation baselines | MNIST softmax regression with same-value attacks, q=4 · MNIST test set, same-value attacks | top-1 accuracy | Helped |
| increase Byzantine workers from q=4 to q=8 | Krum and Median under same-value attacks · MNIST test set, same-value attacks | top-1 accuracy | Hurt |
| use ℓ1-norm RSA under sign-flipping attacks | MNIST softmax regression · MNIST test set, sign-flipping attacks, q=4 and q=8 | top-1 accuracy | Helped |
| use ℓ1-norm RSA instead of Median under sign-flipping attacks | MNIST softmax regression · MNIST test set, sign-flipping attacks | top-1 accuracy | Helped |
| use RSA instead of Median and Krum | MNIST softmax regression · MNIST test set, Gaussian attacks, q=4 and q=8 | top-1 accuracy | Helped |
| use ℓ1-norm RSA instead of Krum | MNIST softmax regression · MNIST test set, heterogeneous data, q=4 | top-1 accuracy | Helped |
| use ℓ1-norm RSA instead of gradient-aggregation baselines | MNIST softmax regression · MNIST test set, heterogeneous data, q=8 | top-1 accuracy | Helped |
| use RSA instead of GeoMed | MNIST softmax regression · MNIST runtime under same-value attacks, q=8 | runtime | Helped |
| use ℓ1-norm RSA instead of Ideal SGD | MNIST softmax regression · MNIST runtime under same-value attacks, q=8 | runtime | Hurt |
| replace ℓ1 regularization with ℓ∞ regularization (ablation) | RSA · MNIST test set, no attacks and same-value attacks with q=8 | top-1 accuracy | Hurt |
| replace ℓ1 regularization with ℓ2 regularization (ablation) | RSA · MNIST test set, no attacks and same-value attacks with q=8 | top-1 accuracy | No clear effect |
Closest papers by meaning
- On the Tension between Byzantine Robustness and No-Attack Accuracy in Distributed Learning (ICML 2025 · similarity 0.92)
- Efficient Federated Learning against Byzantine Attacks and Data Heterogeneity via Aggregating Normalized Gradients (NeurIPS 2025 · similarity 0.91)
- Byzantine-Robust Decentralized Learning via Remove-then-Clip Aggregation (AAAI 2024 · similarity 0.91)
- Self-Driven Entropy Aggregation for Byzantine-Robust Heterogeneous Federated Learning (ICML 2024 · similarity 0.90)
- Fault Tolerant ML: Efficient Meta-Aggregation and Synchronous Training (ICML 2024 · similarity 0.89)
- From Inexact Gradients to Byzantine Robustness: Acceleration and Optimization under Similarity (NeurIPS 2026 · 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.