About me

I am a postdoc in the Department of Computer Science at the University of Manchester, supervised by Dr. Samuel Kaski. I am a member of the Manchester Centre for AI Fundamentals and the ELLIS Society. Much of my research is done in collaboration with the Finnish Center for Artificial Intelligence. I received my PhD in Social and Decision Sciences (concentration in Cognitive Decision Sciences) in 2022 from Carnegie Mellon University, where I was advised by Dr. Daniel Oppenheimer. My dissertation investigated the robustness of Bayesian experimental design to misspecification.

During my PhD, I worked on methods for robust cognitive modeling. I have since extended this work to better understand and address the challenges to robust modeling more broadly, in particular in machine learning. I am particularly interested in the following research areas: model misspecification; experimental design; active learning; Bayesian inference; parameter identifiability and correlation; robust inference; transfer learning. I also maintain a strong interest in the applications of my work to scientific theory and practice, particularly in cognitive science.

Research

Learning from data requires making some assumptions about the structure of the data-generating process, i.e., specifying a model. In practice, that model is often wrong, i.e., misspecified: The prior information used to specify the model was mistaken, the data-generating process changed, and/or the model was a deliberate simplification of the data-generating process. When, and how, can researchers learn useful aspects of the data-generating process with likely misspecified models?

My work takes a data-centric perspective on learning in the sense that I focus on (i) how the structure of the available data affects whether and how much learning occurs, and (ii) the development of methods that alter the structure of one’s data in ways that promote learning (e.g., active learning methods). My recent research is driven by questions like:

• When and how are active learning methods robust to model misspecification?
• How can active learning methods effectively balance the resolution of multiple forms of uncertainty (e.g., between the values of target and nuisance parameters)?
• When and how does increasing the expressivity of a (cognitive or artificial) system improve its ability to learn?
• When predicting in a new environment, when and how can a learning system identify which aspects of what they’ve previously learned will successfully transfer?

Contact

I am always enthusiastic to discuss research or answer questions! You can reach me at firstname (dot) lastname (at) manchester (dot) ac (dot) uk.