Event:

A multi-dimensional stimulus space can be partitioned using different strategies implying different neural tuning-curve shapes. It was recently shown that 3D head-direction - a two-dimensional variable - is encoded in the bat by two sub-populations that form an overcomplete representation (Finkelstein et al., 2015). By analyzing the decoding performance on a simple model we show that these sub-populations are needed to efficiently represent the head-direction in different behaviorally relevant regimes. This suggests that optimal tuning-curve shape can strongly depend on a system’s dynamic variables, which is likely to be relevant to other circuits encoding multi-dimensional stimuli.
Joint work with Arseny Finkelstein, Nachum Ulanovsky and Misha Tsodyks (Weizmann)