Research Centre: Research Centre for Machine Learning
Speaker: Elliot Ludvig
Title: Learning and deciding from replayed experience
Humans and other animals regularly replay their past experiences, as observed in both subjective experience and neural activity. In this talk, I discuss a reinforcement-learning model that can learn from both real and replayed experiences. This enhanced model explains many associative learning phenomena which are troubling for classical computational models, such as latent inhibition and spontaneous recovery. In addition, applied to decision-making, the model assumes that replay need not be veridical, providing a mechanism for how memory biases can influence choice. These biases occur whenever people make decisions based on their past experience, which can lead to both excessive risk-taking and sub-optimal information-seeking. I conclude with some general thoughts about how to improve reinforcement learning with the addition of episodic memories.
Elliot Ludvig is an Associate Professor in the Department of Psychology at the University of Warwick, UK. His research explores how humans and other animals learn to make effective decisions. He completed his Ph.D. in Psychological and Brain Sciences in 2003 at Duke University working on timing in pigeons. In his peripatetic research career, he has also spent time studying the behaviour of humans, animals, and machines at Princeton University, the Technion, the University of Alberta, PsychoGenics Inc., and Rutgers University.
Slides from this seminar can be found here
Share this event
When & where
6.00pm - 7.00pmWednesday 7th November 2018