Research map / Federated learning and privacy / Privacy attacks and secure learning
Attacks against Federated Learning Defense Systems and their Mitigation
Cody Lewis, Vijay Varadharajan, Nasimul Noman
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 |
|---|---|---|---|
| increase adversary share from 10% to 50% (ablation) | FoolsGold · MNIST on-off label flipping | mean ASR: 0.241 -> 0.993 | Hurt |
| increase adversary share from 10% to 30% (ablation) | STD-DAGMM · MNIST on-off label flipping | mean ASR: 0.0253 -> 0.889 | Hurt |
| use stateless Median aggregation instead of history-based defenses (ablation) | Median · MNIST on-off label flipping | mean ASR: 0.0 at 10%, 30% and 50% adversaries | Improved |
| increase adversary share from 30% to 50% (ablation) | Median · MNIST scaled backdoor | mean ASR: 0.126 -> 1.0 | Hurt |
| increase adversary share from 30% to 50% (ablation) | Krum · CIFAR-10 on-off label flipping | mean ASR: 0.0594 -> 0.993 | Hurt |
| increase adversary share from 10% to 50% (ablation) | FoolsGold · MNIST on-off free riding | mean ASR: 0.982 -> 0.0 | Improved |
| increase adversary share from 10% to 50% (ablation) | CONTRA · MNIST on-off free riding | mean ASR: 0.308 -> 0.403 | Hurt |
| start on-off attack in the off (honest) state instead of on (ablation) | FoolsGold · on-off free riding, 10 endpoints, 2 free riders | influence | Hurt |
| use class-imbalanced KDD Cup 99 data (ablation) | FoolsGold · on-off free riding | mean ASR: MNIST 0.982 -> KDD 0.0 at 10% adversaries | Improved |
| increase adversary share from 10% to 50% (ablation) | Median · MNIST good mouthing | mean ASR: 0.968 -> 0.986 | Hurt |
| increase adversary share from 10% to 30% (ablation) | CONTRA · MNIST bad mouthing | mean ASR: 0.335 -> 0.68 | Hurt |
| replace FoolsGold with STD-DAGMM against on-off free riding (ablation) | STD-DAGMM · MNIST on-off free riding, 10% adversaries | mean ASR: 0.982 -> 1.0 | Hurt |
Closest papers by meaning
- Model Poisoning Attacks to Federated Learning via Multi-Round Consistency (CVPR 2025 · similarity 0.87)
- Dual Defense: Enhancing Privacy and Mitigating Poisoning Attacks in Federated Learning (NeurIPS 2024 · similarity 0.87)
- Tracing Back the Malicious Clients in Poisoning Attacks to Federated Learning (NeurIPS 2025 · similarity 0.87)
- FedREDefense: Defending against Model Poisoning Attacks for Federated Learning using Model Update Reconstruction Error (ICML 2024 · similarity 0.86)
- Defending Against Sophisticated Poisoning Attacks with RL-based Aggregation in Federated Learning (AAAI 2025 · similarity 0.86)
- Revamping Federated Learning Security from a Defender's Perspective: A Unified Defense with Homomorphic Encrypted Data Space (CVPR 2024 · similarity 0.86)
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.
