Interpreting wide-band neural activity using convolutional neural networks

Rapid progress in technologies such as calcium imaging and electrophysiology has seen a dramatic increase the size extent of neural recordings. Even so, interpretation this data requires considerable knowledge about nature representation often depends on manual operations. Decoding provides means to infer information content recordings but typically highly processed prior encoding scheme. Here, we developed deep-learning framework able decode sensory behavioral variables directly from wide-band data. The network little user input generalizes across stimuli, behaviors, brain regions, recording techniques. Once trained, it can be analyzed determine elements code that are informative given variable. We validated approach using electrophysiological calcium-imaging rodent auditory cortex hippocampus well human electrocorticography (ECoG) show successful decoding finger movement, spatial behaviors – including novel head direction - raw activity.