Terzi Matteo

Computer Science and Innovation for Societal Challenges, XXXII series
Grant sponsor

GianAntonio Susto
Luciano Serafini, Luciano Gamberini

Project description
My research project is focused in two main parts: the first one is time-series classification and anomaly detection applied to several practical problems (such as activity recognition), which are not been investigated yet or are only partially solved. One of my objectives is to incorporate the dynamics information of the problem at hand within a machine learning algorithm. At this aim, deep learning solutions can provide state-of-art performance in terms of accuracy. However, it is well-known that current DL architectures are not robust with respect to small perturbations applied to the input signal. Currently, the state-of-the-art approach to tackle this issue is adversarial learning which consists on solving a saddle point problem. This approach provides more robust but less accurate solutions. Moreover, the training phase is very slow. Thus, my second part regards the study of robustness of deep-learning algorithm in order to provide solutions that are robust and accurate at the same time.