About

I am a PhD candidate in computing science at the University of Alberta, part of RLAI and Amii, co-supervised by Dale Schuurmans and Marlos C. Machado, and broadly interested in the computational foundations underlying learning from experience: continual learning, reinforcement learning, and meta-learning.

My research seeks to understand the problems faced by agents that continually adapt their behaviour, as they learn from their experience. A central problem that I have studied is loss of plasticity in neural networks, a phenomenon in which performance degrades over the course of learning in non-stationary settings. I developed an optimization perspective on plasticity, showing it can be more effectively sustained by shaping the properties of activation functions and regulating the dynamics of gradient updates. My more recent work investigates a big world setting, where (1) it is suboptimal to stop learning, and (2) this need for continual learning cannot be outscaled. I am working on designing agents that are viable in a big world setting, so that they sustain the complexity of their adaptive behaviour with fixed resources, and grow it with more.

Before my PhD, I completed an MSc in statistics at the University of Alberta (co-supervised by Ivor Cribben and Rohana Karunamuni), and a BMath at the University of Waterloo.

Good news, everyone!

© Alex Lewandowski. Last modified: February 17, 2026. Website built with Franklin.jl and the Julia programming language.