research intern @ FAIR">
Hi! I am a doctoral researcher at Ecole Polytechnique Fédérale de Lausanne (EPFL) since September 2018 in the department of Computer and Communication Sciences. My research focus is understanding the Neural Network magic using a mathematical approach. In particular, I am interested in the overparameterized net landscapes, training dynamics, and both in & out of distribution generalization. I also worked on some classical learning models like random features and kernel methods. I feel very lucky to have my inspiring supervisors Clément Hongler and Wulfram Gerstner.
Before my Ph.D., I studied Electrical and Electronics Engineering and Mathematics (double-major) at Koç University. During my bachelor’s, I spent half a year at the University of California Los Angeles (UCLA) as an exchange student. Even before, I competed in Maths Olympiads and earned two International Mathematical Olympiad (IMO) medals when I was at Izmir Science Highschool. I dance Argentine Tango.
We discuss how the effective ridge reveals the implicit regularization effect of finite sampling in random features. The derivative of the effective ridge tracks the variance of the optimal predictor, yielding an explanation for the variance explosion at the interpolation threshold for arbitrary datasets.