Neural decoding with kernel-based metric learning

摘要: In studies of the nervous system, choice metric for neural responses is a pivotal assumption. For instance, well-suited distance enables us to gauge similarity various stimuli and assess variability repeated stimulus-exploratory steps in understanding how are encoded neurally. Here we introduce an approach where tuned particular decoding task. Neural spike train metrics have been used quantify information content carried by timing action potentials. While number individual neurons exist, method optimally combine single-neuron into multineuron, or population-based, lacking. We pose problem optimizing multineuron other using centered alignment, kernel-based dependence measure. The demonstrated on invasively recorded data consisting both trains local field experimental paradigm consists location tactile stimulation forepaws anesthetized rats. show that optimized highlight distinguishing dimensions response, significantly increase accuracy, improve nonlinear dimensionality reduction methods exploratory analysis.