Yuliang Ma

Research:

In modern industry, CPS (Cyber-PhysicalSystems) are getting more complex due to the increasing number of components inside the systems, such as software, embedded computer, sensor and so on. As a result of increasing structural and behavioral complexity, error/anomaly will inevitably occur in the system. Recently, Deep Learning-based Anomaly Detection (DLAD) methods for CPS have significant advantages in terms of efficiency and accuracy over traditional anomaly detection techniques. However, DLAD methods are black-box techniques and the detection results reported by DLAD methods are not completely dependable because the decision process inside the model is not transparent. Thus, although current DLAD methods have obtained various attractive achievements in the domain of anomaly detection towards CPS, there is still a long way to go to apply these methods into real-world scenarios if the detection results are not completely reliable. Especially when applying DLAD methods for those safety-critical CPS, such as medical robots, human-robot cooperation, mobile robotics, smart factories, users should be aware of the credibility of detection results. And the detected anomalies are also supposed to be interpreted, which can ensure that these safety-critical systems are inspected strictly.

Research portal:

Google Scholar: https://scholar.google.com/citations?user=vfJJEZAAAAAJ&hl=zh-CN

ResearchGate: https://www.researchgate.net/profile/Yuliang-Ma-4