Event Abstract Back to Event Active updating of decision boundaries in rats can be explained using bayesian classifiers It is well established that categorization tasks can be learned through reinforcement, based on reward feedback about the successes and failures of past decisions. Statistical learning theory proposes that active learners use not only reinforcements but also their current estimates of decision uncertainty to set the size of updates. Recent behavioral data from an olfactory categorization task demonstrates that rats show systematic decision biases in trials subsequent to rewarded trials, consistent with ?active updating? (Kepecs et al., Cosyne 2008). Moreover, neural data from the same tasks show that some orbitofrontal cortex neurons encode the likelhood of success for categorization decisions, in a manner consistent with representing decision uncertainty (Kepecs et al., 2008). We apply a recent probabilistic interpretation of Support Vector Machine (SVM) classifers to explain the trial-by-trial updating of the decision boundary in a normative fashion (Tong & Koller, 2000). In Bayesian SVMs (Sollich, 2002), the size of the margin for a sample (distance of the separating hyperplane to the sample) is proportional to the likelihood of that point belonging to a class given the classifier (separating hyperplane). After appropriate normalization, this yields a measure of the posterior variance of the belief state given the current model, i.e., an estimate of confidence about which odor mixture component predominates. Points with high posterior variance are the most informative for updating the decision boundary (Dasgupta et al., 05), and thus the classifier must be preferentially updated with these samples. We used online stochastic gradient descent with an active learning rule to continually update the classifier on a trial-by-trial basis. We simulated the odor categorization experiments and evaluated the performance of our model on synthetic data. The Bayesian SVM model was able to reproduce several key features of the neural and behavioral data. First, the margins of samples from different mixtures show the same characteristic patterns as the firing rates of OFC neurons. Second, our model reproduces the behaviorally observed choice biases shown by rats following a rewarded trial. Finally, the efficacy of the active learning rule is demonstrated in the degree of bias observed following easy and difficult trials-- there is very little change to the boundary following an easy trial. Our results provide a new interpretation of active learning in categorization tasks in terms of margins in Bayesian classifiers. Conference: Computational and systems neuroscience 2009, Salt Lake City, UT, United States, 26 Feb - 3 Mar, 2009. Presentation Type: Poster Presentation Topic: Poster Presentations Citation: (2009). Active updating of decision boundaries in rats can be explained using bayesian classifiers. Front. Syst. Neurosci. Conference Abstract: Computational and systems neuroscience 2009. doi: 10.3389/conf.neuro.06.2009.03.041 Copyright: The abstracts in this collection have not been subject to any Frontiers peer review or checks, and are not endorsed by Frontiers. They are made available through the Frontiers publishing platform as a service to conference organizers and presenters. The copyright in the individual abstracts is owned by the author of each abstract or his/her employer unless otherwise stated. Each abstract, as well as the collection of abstracts, are published under a Creative Commons CC-BY 4.0 (attribution) licence (https://creativecommons.org/licenses/by/4.0/) and may thus be reproduced, translated, adapted and be the subject of derivative works provided the authors and Frontiers are attributed. For Frontiers’ terms and conditions please see https://www.frontiersin.org/legal/terms-and-conditions. Received: 30 Jan 2009; Published Online: 30 Jan 2009. Login Required This action requires you to be registered with Frontiers and logged in. To register or login click here. Abstract Info Abstract The Authors in Frontiers Google Google Scholar PubMed Related Article in Frontiers Google Scholar PubMed Abstract Close Back to top Javascript is disabled. Please enable Javascript in your browser settings in order to see all the content on this page.
Read full abstract