Abstract: How to discriminate visual stimuli based on the activity they evoke in sensory neurons is still an open challenge. To measure discriminability power, we search for a neural metric that preserves distances in stimulus space, so that responses to different stimuli are far apart and responses to the same stimulus are close. Here, we show that Restricted Boltzmann Machines (RBMs) provide such a distance-preserving neural metric. Even when learned in a unsupervised way, RBM-based metric can discriminate stimuli with higher resolution than classical metrics.
TL;DR: RBM-based metric for biological neurons discriminates stimuli with higher resolution than classical metrics.