He has research and work experience in Information Theory and Statistical Physics. More specifically, in the years 2020-21, he worked as a research associate of the Institute of Accelerating Systems and Applications, providing research on the Information Theory of MIMO telecommunication channels.

Research Interests

Thodoris' research interests include Phase Transition Phenomena, Information Theory and Machine Learning. Specifically, his research will focus on areas like loss landscape analysis and the description of learning algorithms on high dimension.

PhD research on "Landscape Complexity for the Empirical Risk of Generalized Linear Models"
Abstract: We calculate, to first exponential order, the number of critical points of certain families of random loss functions for gaussian data with a covariance structure. We focus in the high dimensional asymptotic regime, in which the number of data is proportional to their dimension. This is the only asymptotic regime in which the problem is non-trivial and consequently it is the relevant regime for modern data science.