Recording
Speaker
Christopher Summerfield
Bio
Christopher Summerfield is a Fellow by special election and principal investigator at the Summerfield lab which conducts research into how humans make decisions. Chris Summerfield was trained in psychology and neuroscience at University College London, Columbia University (New York), and the École normale supérieure (Paris). He is a Professor of Cognitive Neuroscience in the Department of Experimental Psychology, where he heads a lab focused on understanding the computational mechanisms by which humans make decisions, and how these processes are implemented in the brain. His work, which involves a combination of computer simulations, behavioural testing, and functional brain imaging, is funded by a grant from the European Research Council, the Wellcome Trust, and the National Institute of Health. Recently, he accepted the position as the director at UK AI Safety Institute.
Abstract
Connectionist models have made a comeback as theories of brain function. Most advocates of this view have compared the structure of representations in biological and artificial neural networks, and argued that they are similar. However, neural networks learn with stereotyped dynamics which in some cases can be modelled exactly. In my talk, I will discuss projects in which we compare the learning dynamics in humans and deep networks during hierarchical category learning, task switching, dual task learning, and in-context (or “meta-“) learning. In each of these cases, we find striking commonalities between the dynamics of learning in deep networks, and those in our human participants.