Emergence of functional and structural properties of the head direction system by optimization of recurrent neural networks

• DOI: 10.48550/arxiv.1912.10189
• Publication Date: 2019-01-01
• Authors: Christopher J. Cueva; Peter Y. Wang; Matthew Chin; Xue-Xin Wei
• Keywords: Biological neural network; Neurophysiology; Compass

Recent work suggests goal-driven training of neural networks can be used to model activity in the brain. While response properties neurons artificial bear similarities those brain, network architectures are often constrained different. Here we ask if a recover both representations and, architecture is unconstrained and optimized, anatomical circuits. We demonstrate this system where connectivity functional organization have been characterized, namely, head direction circuits rodent fruit fly. trained recurrent (RNNs) estimate through integration angular velocity. found that two distinct classes observed system, Compass Shifter neurons, emerged naturally as result training. Furthermore, analysis in-silico neurophysiology revealed structural mechanistic between system. Overall, our results show optimization RNNs task recapitulate structure function biological circuits, suggesting study brain at level organization.