Interpreting wide-band neural activity using convolutional neural networks

Abstract

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

Predictive Maps in Rats and Humans for Spatial Navigation

“…Further, the points where agents switch between different options may be able to predict where rats and humans would pause in the maze 72 . More broadly, new approaches to RL 109 and deep learning methods may provide new ways to examine navigation 40, 110 , as well as integrating our approach with biologically-inspired network models that seek to explain neural dynamics during navigation 111, 112 .…”

“…Balaguer et al, 2016). More broadly, new approaches to RL 106 and deep learning methods may provide new ways to examine navigation 40,107 , as well as integrating our approach with biologically-inspired network models that seek to explain neural dynamics during navigation 108,109 .…”

Predictive maps in rats and humans for spatial navigation

“…Furthermore, the points where agents switch between different options may be able to predict where rats and humans would pause in the maze. 73 More broadly, new approaches to RL 111 and deep learning methods may provide new ways to examine navigation 41,112 as well as integrating our approach with biologically inspired network models that seek to explain neural dynamics during navigation. 113,114 Exploring the neural substrates of a predictive map Recent neuroimaging in humans has shown that activity in hippocampal and connected regions tracks the modeled parameters from a SR. 63,64,66,68,78 Convergent evidence in rodents suggests that the place-cell activity in the dorsal cornu Ammonis 1 (CA1) of rodents may operate as a SR. 67,68,71,115,116 Our protocol would allow for evidence from both rodent and human data to be integrated within a single framework to consider how patterns in the data may interrelate across species and in relation to the parameters from RL-modeled agents.…”

Model discovery to link neural activity to behavioral tasks

“…MINE specifically supports an encoding view, inspecting which information is encoded by the neurons of interest. Artificial neural networks have however also been used from a decoding perspective, probing what in-formation can be gleaned about ongoing behavior from neural activity 51,52 . Here we restricted the question of information encoding to individual neurons.…”