Phillip Rieger (Technical University of Darmstadt), Thien Duc Nguyen (Technical University of Darmstadt), Markus Miettinen (Technical University of Darmstadt), Ahmad-Reza Sadeghi (Technical University of Darmstadt)

Federated Learning (FL) allows multiple clients to collaboratively train a Neural Network (NN) model on their private data without revealing the data. Recently, several targeted poisoning attacks against FL have been introduced. These attacks inject a backdoor into the resulting model that allows adversary-controlled inputs to be misclassified. Existing countermeasures against backdoor attacks are inefficient and often merely aim to exclude deviating models from the aggregation. However, this approach also removes benign models of clients with deviating data distributions, causing the aggregated model to perform poorly for such clients.

To address this problem, we propose emph{DeepSight}, a novel model filtering approach for mitigating backdoor attacks. It is based on three novel techniques that allow to characterize the distribution of data used to train model updates and seek to measure fine-grained differences in the internal structure and outputs of NNs. Using these techniques,
DeepSight can identify suspicious model updates. We also develop a scheme that can accurately cluster model updates. Combining the results of both components, DeepSight is able to identify and eliminate model clusters containing poisoned models with high attack impact. We also show that the backdoor contributions of possibly undetected poisoned models can be effectively mitigated with existing weight clipping-based defenses. We evaluate the performance and effectiveness of DeepSight and show that it can mitigate state-of-the-art backdoor attacks with a negligible impact on the model's performance on benign data.

View More Papers

PMTUD is not Panacea: Revisiting IP Fragmentation Attacks against...

Xuewei Feng (Tsinghua University), Qi Li (Tsinghua University), Kun Sun (George Mason University), Ke Xu (Tsinghua University), Baojun Liu (Tsinghua University), Xiaofeng Zheng (Institute for Network Sciences and Cyberspace, Tsinghua University; QiAnXin Technology Research Institute & Legendsec Information Technology (Beijing) Inc.), Qiushi Yang (QiAnXin Technology Research Institute & Legendsec Information Technology (Beijing) Inc.), Haixin Duan…

Read More

CFInsight: A Comprehensive Metric for CFI Policies

Tommaso Frassetto (Technical University of Darmstadt), Patrick Jauernig (Technical University of Darmstadt), David Koisser (Technical University of Darmstadt), Ahmad-Reza Sadeghi (Technical University of Darmstadt)

Read More

FANDEMIC: Firmware Attack Construction and Deployment on Power Management...

Ryan Tsang (University of California, Davis), Doreen Joseph (University of California, Davis), Qiushi Wu (University of California, Davis), Soheil Salehi (University of California, Davis), Nadir Carreon (University of Arizona), Prasant Mohapatra (University of California, Davis), Houman Homayoun (University of California, Davis)

Read More

MIRROR: Model Inversion for Deep LearningNetwork with High Fidelity

Shengwei An (Purdue University), Guanhong Tao (Purdue University), Qiuling Xu (Purdue University), Yingqi Liu (Purdue University), Guangyu Shen (Purdue University); Yuan Yao (Nanjing University), Jingwei Xu (Nanjing University), Xiangyu Zhang (Purdue University)

Read More