Talk page
Title:
Temporal stimulus segmentation by reinforcement learning in populations of spiking neurons
Speaker:
Abstract:
Stimulus identification is the process of picking out a particular stimulus among many other stimuli which may be present in the environment, for the purpose of performing a task. In the most interesting yet demanding scenario, the problem amounts to extracting action-relevant segments out of a noisy input stream, thus also involving the extrapolation of the beginning and the end of relevant stimulus features, not known a priori. In a typical framework, the identity and timing of the relevant stimuli are known to the learning agent. Departing from this view, I introduce an autonomous learning system able to identify action-relevant stimuli without prior knowledge of them, while learning to ignore stimuli that are not behaviorally relevant. The model is based on ensembles of spiking neurons able to handle spatiotemporal patterns of spike trains reminiscent of those recorded in alert animals performing decision tasks. It uses a biologically plausible learning rule for maximizing the average reward obtained for correct decisions taken at the right time. Performance of the model on surrogate and cortical datasets is robust to stimulus noise and multiple coding strategies, providing an example of a spiking network model able to solve multi-choice decision tasks in the absence of prior information on the relevance and timing of input stimuli.
Link:
Workshop: