Local perturbation-based black-box federated learning attack for time series classification

Shengbo Chen, Jidong Yuan, Zhihai Wang, Yongqi Sun

Abstract: Highlights•Our method is the first to propose a federated learning attack on time series.•We introduce a gradient-free, black-box federated learning backdoor attack method.•We use local perturbations to make the method highly stealthy on time series data.•Our attack outperforms other baselines and successfully bypasses defense methods.