This contains the simulation code to our paper, https://doi.org/10.7554/eLife.50804.

List of files perceptron code to produce the data for Figure 1 and can be edited to produce the data for the other figures related to the perceptron.

• cperc.m: Octave file for simulating perceptron learning
• runc.cc: For efficiency, the actual learning is done in this C code.
• Makefile: compiles runc.cc

The Octave script runs multi-threaded using pararrayfun; replacing it with arrayfun will run single threaded. Conversion to Matlab should be straightforward and is left as an exercise.

List of multi-layer codes to produce Figure 4. Each panel in the figure has a specific code to generate its data, which then is plotted by a respective "MakePlot" Python code.

• learning_rate.py: generating data for panel (a), and its plotting code is learning_rate_MakePlot.py
• energy_vs_accuracy.py: for (b), the plotting code is energy_vs_accuracy_MakePlot.py
• nHidden.py: for (c), the plotting code is nHidden_MakePlot.py
• KleinFunction.py: this contains the function that conducts the back-propagation training, and is called by the Python codes above.

List of codes to produce the energy cost of learning by using D'Souza et al. (2010) learning rule (Figure 2-Figure 1 Supplement):

• standalone_STDP.py: generating data
• standaloneFunction.py: contains functions that are called by standalone_STDP.py
• makeplot.py: for analysing the data and plotting the result.