MF-GC-network-backprop-data

This repo contains pre-simulated data necessary to reproduce all major figures for Cayco-Gajic, Clopath & Silver 2017, "Sparse synaptic connectivity is required for decorrelation and pattern separation in feedforward networks."

Note that throughout the data, “_rX” indicates correlation radius (sigma) of X, so that r0 indicates independent inputs, r5 indicates spatial correlations of 5 microns, etc. Similarly, “_shuff” indicates that the GC activity patterns have been partially shuffled, and in results_bp_th, “_rX_Y” indicates sigma of X and threshold of Y.

Includes:

• In network_structures:
  • Generated network connectivities and MF rosette/ GC positions (e.g. GCLconnectivity_4.pkl)
• In input_statistics:
  • Generated input statistics for varying spatial correlations (e.g. mf_patterns_r0.mat)
• In biophysical model:
  • Initialized parameters for cluster array job (e.g. params_file.pkl)
  • Variance and covariance of pre-simulated activity patterns (e.g. data_r0/grc_cov_biophys_r0.mat)
  • Population sparseness of pre-simulated activity patterns: (e.g. data_r0/grc_spar_biophys_r0.mat)
  • Error from backpropagation training (e.g. data_r0/grc_bp_biophys_r0.mat)
• In analytical model:
  • Error from backpropagation training (e.g. results_bp/grc_toy_r0.mat)

This repo only contains data. For necessary scripts, see: https://github.com/SilverLabUCL/MF-GC-network-backprop-public

Warning before you clone: This repo is ~2Gb.