Current students

Detecting changes in the distribution of a process generating a stream of signals or other high-dimensional data is a very important problem with several industrial applications. As a relevant example, I have investigated how to employ advanced sequential monitoring techniques to strengthen side-channel cryptographic attacks, which are based on monitoring the power consumption and execution times of cryptographic devices. This research was done in collaboration with STMicroelectronics. I am also addressing the more fundamental research problem of extending sequential monitoring in the context of multivariate and potentially high-dimensional data streams. In particular, the change-detection solution I have proposed to strengthen side-channel attacks, is limited to 1-dimensional signals by some theoretical constraints. Thus, despite being very powerful, it cannot be applied to multivariate data streams. I am currently investigating an extension of QuantTree, which is a very powerful change-detection method for multivariate data but it is limited to  batch-wise monitoring, to a truly sequential monitoring scheme. Future work concerns change/anomaly detection problems for very high-dimensional data (such as wafer defect maps from silicon manufacturing) that have necessarily to combine data-driven models (deep learning). In this direction, I will investigate new learning strategies to improve change/anomaly detection performance.