Atli Kosson

Short Bio: I am a fourth-year computer science PhD student at EPFL in Switzerland🇨🇭 supervised by Professor Martin Jaggi. Previously, I worked as an Autopilot Engineer at Tesla optimizing large-scale neural network training and as an ML Research Engineer at Cerebras on efficient training algorithms. I did my Master’s at Stanford University and my undergrad at the University of Iceland 🇮🇸 where I grew up.

Research Focus: I am broadly interested in improving our understanding of neural network architectures, internal representations, and optimization. Although my focus is on understanding, I lean towards the “top-down” style of experimentation and applied analysis instead of a “bottom-up” theoretical approach. Currently, I am investigating how and why optimization tricks like weight decay and adaptive learning rates aid training.