Michael Mozer received a Ph.D. in Cognitive Science at the University of California at San Diego in 1987.  Following a postdoctoral fellowship with Geoffrey Hinton at the University of Toronto, he joined the faculty at the University of Colorado at Boulder and is presently an Professor in the Department of Computer Science and the Institute of Cognitive Science.  He is secretary of the Neural Information Processing Systems Foundation, has served as Program Chair and General Chair at NIPS and as chair of the Cognitive Science Society. He is interested in human-centric artificial intelligence, which involves designing machine learning methods that leverage insights  from human cognition, and building software tools to optimize human performance using machine learning methods.

AG 1, AG 2, AG 3, AG 4, AG 5, SWS, RG1, MMCI   Info

AG Audience

English

60 Minutes

Kaiserslautern

We are developing methods to improve human learning and performance on challenging perceptual categorization tasks, e.g., bird species identification, diagnostic dermatology. Our approach involves inferring psychological embeddings -- internal representations that individuals use to reason about a domain. Using predictive cognitive models that operate on an embedding, we perform surrogate-based optimization to determine efficient and effective means of training domain novices as well as amplifying an individual's capabilities at any stage of training. Our cognitive models leverage psychological theories of: similarity judgement and generalization, contextual and sequential effects in choice, attention shifts among embedding dimensions.  Rather than searching over all possible training policies, we focus our search on policy spaces motivated by the training literature, including manipulation of exemplar difficulty and the sequencing of category labels. We show that our models predict human behavior not only in the aggregate but at the level of individual learners and individual exemplars, and preliminary experiments show the benefits of surrogate-based optimization on learning and perform ance.

This work was performed in close collaboration with Brett Roads at University
College London.

Susanne Girard

--email hidden

Yes

Saarbrücken

E1 5

029

passcode not visible

logged in users only