EpigeneSys

Epigenesis of the Global Neuronal Workspace.

Multi-Level Development of Cognitive Abilities in an Artificial Neural Network

Dates: 2018-2022

Authors: Konstantin Volzhenin, Jean-Pierre Changeux, Guillaume Dumas

• GNW: Scripts for the final paper reproducing the results of the paper.

From Local Hebbian to Global Reinforcement Learning

Dates: 2015-2018

Authors: Guillaume Dumas, Jean-Stanislas Denain, Valentin Villecroze, Camille Démarre, Gauthier Guinet, Paul Jacob, Anne-Sophie Migeon, Jean-Pierre Changeux

Poster presented at the symposium "Neural networks - from brains to machines and vice versa" on the 11th - 12th October 2018 at the Institut Pasteur, Paris, France.

Implementation in Brian 2 of the Izhikevich's 2007 synapse model

• SynapseModel/STDP_Reward: Brian 2 implementation of Izhikevich's synapse model as described in his 2007 paper (see equation below).

Plots the evolution of the model parameters for a simple two-neuron network:

Synapse reinforcement

• ReinforceSynapse/ReinforceSynapse: Test of the first experiment in Izhikevich's 2007 paper. In a large network of neurons, release dopamine when the synapse between two given neurons is activated. The goal is to check whether this synapse is selectively reinforced.

Conditioning

• Conditioning/SimpleConditioning: Test of the second experiment in Izhikevich's 2007 paper. In a network composed of multiple small groups of neurons, release dopamine whenever neurons in the first group spike.

Other early sandboxes

• MultiColumns/GNW: Architecture for a structured multi-column network in Brian.
• XOR/XOR: Sandbox for the XOR function.
• MultiColumns/Digits: Sandbox for MNIST digits recognition.