Dendritic balance enables local learning of efficient representations in networks of spiking neurons

These files accompany the results obtained in Dendritic balance enables local learning of efficient representations in networks of spiking neurons. For further details please refer to the manuscript posted on PNAS.

How to recreate results

Figure 3

Run julia mnist_().jl, where () is to be replaced to execute the desired file.

To plot results run python plot_mnist.py ../../Fig3_mnist_all_weights_decay/logs/() ../../Fig3_mnist_somatic/logs/(), where () is to be replaced with the desired folder.

Figure 4

B

Run julia B_bars_().jl $i, where $i ranges from 1-1050. We recommend only using the analytic and somatic implementations, as the all weights implementations are slow.

To plot the overview results first concat the resulting losses into one file cat results* > cresults.txt, in the log folders. Then run python B_plot_bars.py.

D

Run julia D_scenes_rate_scan.jl $i and julia D_scenes_rate_scan_somatic.jl $i, where $i ranges from 1-9.

To plot the overview results run python D_plot_scenes_comparison.py ../../Fig4_scenes_rate_scan/logs/ ../../Fig4_scenes_rate_scan_somatic/logs/.

Figure 5

A

Run julia A_scenes_timestep_scan.jl $i and julia A_scenes_timestep_scan_somatic.jl $i, where $i ranges from 1-72.

To plot the overview results run python A_plot_scenes_comparison.py ../../Fig5_scenes_somatic_timestep_scan/logs/ ../../Fig5_scenes_somatic_timestep_scan_somatic/logs/.

B

Same as in figure 3.

Requirements

The results in this paper were created using Julia 1.3.1 and Python 3.6 with matplotlib, numpy and h5py. To run the experiments using natural images, please download IMAGES.mat from http://www.rctn.org/bruno/sparsenet/ and place it into src/input_generation/scenes/. To run the experiments using speech data, please download speech.mat from https://github.com/machenslab/spikes/tree/master/UnsupervisedLearning/Figure5 and place it into src/input_generation/speech/

Julia Packages

pkgs = ["BSON",
"Dates",
"DelimitedFiles",
"FileIO",
"HDF5",
"ImageMagick",
"Images",
"InteractiveUtils",
"JSON",
"LinearAlgebra",
"MAT",
"MLDatasets",
"MultivariateStats",
"Plots",
"Profile",
"ProgressMeter",
"PyPlot",
"SparseArrays"]

using Pkg

Pkg.add(pkgs)