Cognitive Lunch Abstract for April 8

The next Cognitive Lunch will be held Wednesday, April 8
Time: 12:10-1:10
Place: Psychology Conference Room (room 128)

"A Bayesian Framework for Active Learning" will be presented by Stephen E. Denton and John K. Kruschke, Department of Psychological and Brain Sciences, Indiana University.

Abstract:
Bayesian approaches to learning involve incrementally updating degrees of belief across a space of hypotheses whenever an observer passively observes a stimulus and outcome. But these approaches also provide a framework for models of active learning, because uncertainty across beliefs can be evaluated and thus the expected uncertainty reduction for candidate stimuli can be computed. Bayesian models of active learning predict that an active learner would select the stimulus for which expected uncertainty across all hypotheses is minimized. Our research contrasts four possible hypotheses spaces consisting of two simple cue-combination models and two possible priors. To tease apart the models, we performed an automated search of associative learning structures for which the models make maximally different predictions. Human participants were tested on these same learning structures in the context of an allergy diagnosis task. At various points in training we asked which cues they would find most informative to learn about; i.e., we assessed the participant's active learning preferences. The model and prior combinations that best mimic human active learning will be discussed.