Harnessing Cortical Geometry, Wiring, and Function as Inductive Biases for Recurrent Neural Networks

This repository contains the implementation for the research paper "Harnessing Cortical Geometry, Wiring, and Function as Inductive Biases for Recurrent Neural Networks". The work explores how incorporating neuronal constraints from the MICrONS dataset, can enhance the performance and realism of recurrent neural networks (RNNs) on decision-making tasks.

Repository Structure

neuro-constrained-RNN/
├── README.md                          # This file
├── neuro-constrained-RNN.ipynb        # Main experimental notebook
├── DATA.zip                           # Preprocessed MICrONS dataset
└── INF.md                             # Dataset information (19,178 neurons)