Konstantinos Christopher Tsiolis
Hello there! I am a PhD candidate at the Department of Statistical Sciences at the University of Toronto and the Vector Institute under the supervision of Murat Erdogdu.
I am broadly interested in the rigorous theoretical analysis of machine learning methods. The primary objective of my current research is to characterize learning dynamics under stochastic gradient descent (SGD) using tools from high-dimensional statistics, with an eye towards feature learning and scaling laws.
Previously, I graduated with an MSc in Mathematics and Statistics and a BSc in (Honours) Mathematics and Computer Science at McGill University in my beautiful hometown of Montreal. During my MSc, I wrote a thesis on contrastive self-supervised learning methods under the supervision of Adam Oberman. During my BSc, I worked on problems in natural language processing (NLP) under the supervision of Jackie Cheung at Mila and the McGill Reasoning and Learning Lab.
Contact: kc (dot) tsiolis (at) mail (dot) utoronto (dot) ca
News
- September 2026: My new preprint “SGD in Multiclass Logistic Regression: Sequential Learning and Scaling Laws” is out now! This is joint work with Denny Wu, Christos Thrampoulidis, and Murat Erdogdu.
- June 2026 I gave a contributed talk on the theory of scaling laws at the Stein’s Method Meets Statistical Learning workshop, held at the Banff International Research Station (BIRS).
- December 2025: I presented my work “From Information to Generative Exponent: Learning Rate Induces Phase Transitions in SGD” at NeurIPS 2025 in San Diego. This is joint work with Alireza Mousavi-Hosseini and Murat Erdogdu.
- November 2025: I am very grateful to have received the Department of Statistical Sciences Student Leadership Award!