Conference on Cognitive Computational Neuroscience

Distinguishing between complex nonlinear neural time-series patterns is a challenging problem in neuroscience. Accurately classifying different patterns could be useful for a wide variety of applications, e.g. detecting seizures in epilepsy. On the one hand, there are simple distance metrics, which can be computed quickly, but do not yield accurate classifications; on the other hand, deep supervised approaches offer high accuracy but are training data intensive. We introduce a reservoir-based tool, state tracker (TRAKR), which provides the high accuracy of deep supervised methods while preserving the benefits of simple distance metrics in being applicable to single examples of training data. We apply TRAKR to a benchmark dataset – permuted sequential MNIST (psMNIST) – and show that it achieves high accuracy. We also apply TRAKR to electrocorticography (ECoG) data from the macaque orbitofrontal cortex (OFC) and find that TRAKR can distinguish 3 behaviorally relevant epochs (rest, choice and reward seeking periods) accurately. In both cases, TRAKR performs on par with deep networks and better than simple distance metrics. These results demonstrate that TRAKR could be a viable alternative in the classification of neural time-series data, offering the potential to generate new insights into the information encoded in neural circuits from single-trial data.