Biological learning and memory from a theoretical perspective

Add to your list(s) Download to your calendar using vCal

• Mate Lengyel (Engineering Department)
• Wednesday 21 May 2008, 02:35-03:20
• Centre for Mathenatical Sciences, Meeting Room 2.

If you have a question about this talk, please contact Danielle Stretch.

This talk has been canceled/deleted

The sole purpose of having memories about the past is to aid adaptive behaviour in the future. I will present recent work formalizing this normative idea mathematically and employing it to understand two different memory systems in the brain. Learning about the joint statistics of multiple environmental variables is the basis for making valid predictions. Traditionally, learning has been described as being achieved through storing pairwise associations between variables. However, from a theoretical perspective, this seems to be clearly sub-optimal: it leads to representing only second-order correlations of the available information while the statistics of our environments are much more richly structured. Bayesian statistics offers a principled solution to selecting the optimal complexity of a representation. Thus, we investigated whether human learning is best described by simple associative or more sophisticated Bayesian learning in a visual chunk learning paradigm. We found that human subjects can learn about the statistics of visual stimuli in a highly efficient way, close to the optimum defined by Bayesian model comparison, which surpasses what was predicted by conventional pairwise associative theories. In light of the efficiency of this form of ‘semantic’ learning, it is puzzling why we have other memory systems, such as the one that stores individual autobiographical episodes? To answer this question, rather than considering single isolated predictions, we have studied the use of memories for sequential decision making, when feedback (in the form of rewards) is delayed. We show that under specific, and behaviourally relevant, conditions a system storing episodes can best a system using ‘semantic’ memories, storing the sufficient statistics of the environment, because its relative inefficiency to represent information about the environment is compensated for by the accuracy of decisions that can be based on it. The brain has a unique capacity to learn continuously about the environment and to use this knowledge flexibly to make predictions and guide its future decisions. Mathematical theories can provide a powerful tool for dissecting the impressive complexities underlying this remarkable feat.

This talk is part of the One Day Meeting - Fourth Annual Symposium of the Cambridge Computational Biology Institute series.

This talk is included in these lists:

This talk is not included in any other list

Note that ex-directory lists are not shown.