Keynote Speakers
Peter Dayan, Director of the Gatsby Computational Neuroscience Unit, University College London
Betwixt fast and slow: Integrating model-free and model-based decision-making
Behavioural and neural evidence reveals a retrospective, model-free or habitual process that caches returns previously garnered from available choices, and a prospective, model-based or goal-directed one that putatively relies on mental simulation of the environment. There is much current interest in understanding how these faster and slower systems are integrated to take advantage of the beneficial computational properties of each. I will discuss a recent study looking theoretically and empirically at the incorporation of model-free values into model-based calculations. This is joint work with Mehdi Keramati, Peter Smittenaar and Ray Dolan.

Randy Gallistel, Distinguished Professor Emeritus, Rutgers University
Information Theory and Stochastic Model Selection in Associative Learning and Memory
Two information-theoretic principles, maximum entropy, and minimum description length found a computational model of associative learning that explains cue competition (assignment of credit), response timing, and the parametric invariances. State cues and point cues are linked, respectively, to two stochastic distributions, the exponential and the BernoulliGauss. The stochastic model selected by the computational model specifies the relative code lengths for the most efficient encoding of the data and best predicts the data not yet seen. Its hazard function predicts the timing of conditioned behavior. The minimum-description-length approach to stochastic model selection (Rissanen 1999) enables the computational model to find the stochastic model that maximally compresses the data and best predicts the future.

Joe Houpt, William K. Estes Early Career Award Winner, Wright State University