Multi-metrics adaptively identifies backdoors in Federated learning
Authors:
- 1 October 2023
- conference paper
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 4629-4639
Abstract
The decentralized and privacy-preserving nature of federated learning (FL) makes it vulnerable to backdoor attacks aiming to manipulate the behavior of the resulting model on specific adversary-chosen inputs. However, most existing defenses based on statistical differences take effect only against specific attacks, especially when the malicious gradients are similar to benign ones or the data are highly non-independent and identically distributed (non-IID). In this paper, we revisit the distance-based defense methods and discover that i) Euclidean distance becomes meaningless in high dimensions and ii) malicious gradients with diverse characteristics cannot be identified by a single metric. To this end, we present a simple yet effective defense strategy with multi-metrics and dynamic weighting to identify backdoors adaptively. Furthermore, our novel defense has no reliance on predefined assumptions over attack settings or data distributions and little impact on benign performance. To evaluate the effectiveness of our approach, we conduct comprehensive experiments on different datasets under various attack settings, where our method achieves the best defensive performance. For instance, we achieve the lowest backdoor accuracy of 3.06% under the most difficult Edge-case PGD, showing significant superiority over previous defenses. The experiments also demonstrate that our method can be well-adapted to a wide range of non-IID degrees without sacrificing the benign performance.Keywords
All Related Versions
Funding Information
- Research and Development
This publication has 24 references indexed in Scilit:
- More than Encoder: Introducing Transformer Decoder to Upsample2022
- Weakly supervised object localization with soft guidance and channel erasing for auto labelling in autonomous driving systemsISA Transactions, 2022
- Swin Transformer V2: Scaling Up Capacity and Resolution2022
- OoD-Bench: Quantifying and Understanding Two Dimensions of Out-of-Distribution Generalization2022
- Collaborative Unsupervised Visual Representation Learning from Decentralized Data2021
- Invisible Backdoor Attack with Sample-Specific Triggers2021
- Towards Efficient and Privacy-Preserving Federated Deep Learning2019
- On the Surprising Behavior of Distance Metrics in High Dimensional Space2001
- When Is “Nearest Neighbor” Meaningful?1999
- Gradient-based learning applied to document recognitionProceedings of the IEEE, 1998