BARS: Local Robustness Certification for Deep Learning based Traffic Analysis Systems

• DOI: 10.14722/ndss.2023.24508
• Publication Date: 2023-03-12T13:00:37Z
• Authors: Kai Wang; Zhiliang Wang; Dongqi Han; Wenqi Chen; Jiahai Yang; Xingang Shi; Xia Yin
• Keywords: Robustness

Deep learning (DL) performs well in many traffic analysis tasks.Nevertheless, the vulnerability of deep weakens real-world performance these analyzers (e.g., suffering from evasion attack).Many studies recent years focused on robustness certification for DL-based models.But existing methods perform far perfectly domain.In this paper, we try to match three attributes systems at same time: (1) highly heterogeneous features, (2) varied model designs, (3) adversarial operating environments.Therefore, propose BARS, a general framework based boundary-adaptive randomized smoothing.To obtain tighter guarantee, BARS uses optimized smoothing noise converging classification boundary.We firstly Distribution Transformer generating noise.Then optimize noise, some special distribution functions and two gradient searching algorithms shape scale.We implement evaluate practical systems.Experiment results show that can achieve guarantee than baseline methods.Furthermore, illustrate practicability through five application cases quantitatively evaluating robustness).