Research map / Federated learning and privacy / Privacy attacks and secure learning
Chameleon: Adapting to Peer Images for Planting Durable Backdoors in Federated Learning
Yanbo Dai, Songze 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 |
|---|---|---|---|
| Use Chameleon contrastive adaptation and classifier projection instead of prior backdoor attacks | Federated visual backdoor training · CIFAR100 bicycle with black background | 50%-Lifespan: 229 -> 661 | Improved |
| Use Chameleon instead of prior attacks | Federated visual backdoor training · CIFAR10 car with striped background | 50%-Lifespan: 283 -> 1106 | Improved |
| Use Chameleon instead of prior attacks | Federated visual backdoor training · CIFAR10 southwest airline edge-case | 60%-Lifespan: 339 vs 349 | Mixed |
| Use Chameleon rather than Baseline | CIFAR100 bicycle with black background under weakly DP and NCD · CIFAR100 | 50%-Lifespan: 150 vs - | Improved |
| Use Chameleon rather than Neurotoxin | CIFAR100 bicycle with black background under Krum and NCD · CIFAR100 | 50%-Lifespan: 53 -> 72 | Improved |
| Add Chameleon’s peer-image-aware training | CIFAR100 bicycle with black background in FedProx · FedProx, Dirichlet α=0.9 | 50%-Lifespan: 472 -> 1522 | Improved |
| Increase pattern diffusion from p_d=0.1 to p_d=0.5 (ablation) | Chameleon · CIFAR10 Type-2 pixel-pattern backdoor | 50%-Lifespan: 833(50%) -> 720(50%) | Hurt |
| Increase β from 3 to 12 (ablation) | Chameleon pixel-pattern backdoor · Type-2 pixel-pattern, p_d=0.5 | Backdoor durability | Improved |
| Vary β, with the strongest result at an intermediate β=4 (ablation) | Chameleon semantic backdoor · CIFAR10 car-with-racing-stripes | Backdoor durability: β=4 | Mixed |
| Remove interferers from subsequent benign-client datasets (ablation) | Backdoor trained with model replacement · CIFAR10 car-to-bird semantic backdoor | Backdoor accuracy persistence | Improved |
| Remove facilitators from subsequent benign-client datasets (ablation) | Backdoor trained with model replacement · CIFAR10 car-to-bird semantic backdoor | Backdoor durability | Hurt |
| Train using only poisoned images rather than mixing in benign images (ablation) | Malicious local backdoor model · CIFAR10 car-to-bird semantic backdoor | Backdoor lifespan | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- Neurotoxin — Neurotoxin: Durable Backdoors in Federated Learning (ICML 2022) · better on 4 of 4 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Resisting Backdoor Attacks in Federated Learning via Bidirectional Elections and Individual Perspective (AAAI 2024 · similarity 0.89)
- Beyond Traditional Threats: A Persistent Backdoor Attack on Federated Learning (AAAI 2024 · similarity 0.88)
- Infighting in the Dark: Multi-Label Backdoor Attack in Federated Learning (CVPR 2025 · similarity 0.87)
- MARS: A Malignity-Aware Backdoor Defense in Federated Learning (NeurIPS 2025 · similarity 0.87)
- SPMC: Self-Purifying Federated Backdoor Defense via Margin Contribution (ICML 2025 · similarity 0.87)
- Label-Free Backdoor Attacks in Vertical Federated Learning (AAAI 2025 · similarity 0.87)
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.
