Exploring Adversarial Robustness of Deep State Space Models Biqing Qi

Deep State Space Models (SSMs) have proven effective in numerous task scenarios but face significant security challenges due to Adversarial Perturbations (APs) in real-world deployments. Adversarial Training (A T) is a mainstream approach to enhancing Adversarial Robustness (AR) and has been validated on various traditional DNN architectures. However, its effectiveness in improving the AR of SSMs remains unclear. While many enhancements in SSM components, such as integrating Attention mechanisms and expanding to data-dependent SSM pa-rameterizations, have brought significant gains in Standard Training (ST) settings, their potential benefits in A T remain unexplored. To investigate this, we evaluate existing structural variants of SSMs with A T to assess their AR performance.