Kavli Affiliate: Terrence Sejnowski
| Authors: Yusi Chen, Huanqiu Zhang and Terrence J Sejnowski
| Summary:
We linked the temporal precision of neural encoding to the implementation of cognitive functions through predictive sequence learning. The hippocampus of rodents receives sequences of sensory inputs from the cortex during exploration and then encodes the sequences with millisecond precision despite inter-regional transmission delays. Our study linked such temporal precision to the cognitive functions of hippocampus in a self-supervised recurrent neural network that was trained to predict its next input. The model exhibited localized place cells and experimentally observed features such as one-shot learning, replay and phase precession. We tested and confirmed the assumption that area CA3 is a predictive recurrent autoencoder by analyzing the spike coupling between simultaneously recorded neurons in hippocampal subregions. These results imply that the place field activity of neurons in area CA1 report temporal prediction error, which decays with familiarity.