CS 229: Machine Learning Classifying Modality of Pain from Calcium Imaging Florescence data

Motivation. When you are bitten by a mosquito, you know that the bump itches. When you burn your hand on a match, you know that the match was hot. This ability to discriminate sensory modalities is present, despite the fact that these stimuli activate overlapping sets of peripheral receptors, and travel to your brain generally along the same paths, utilizing overlapping sets of neurons [1]. How, or where, this distinction emerges, remains unclear in the neuroscience literature. To overcome this limitation, we designed a machine learning approach, whereby we built a classifier to determine, purely from neural data, what modality of stimulus we present to an animal. The logic is that if the brain region from whence we record neural data is actively involved in task of detecting the different stimuli modalities, are decoder will be able to easily distinguish between them. This approach is similar to one often taken in both visual and decision making literature [2].