This is a python implementation of Bonaiuto and Arbib's scientific paper Extending the mirror neuron system model, II: what did I just do? A new role for mirror neurons.
It is an attempt to reproduce the results for teaching purposes.
Requires python 2.7+.
The neural network simulating the mirror system is implemented with pytorch.
Additional libraries can be install through pip, using
pip install -r requirements.txt
The parameters of the Augmented Competitive Queue (ACQ) can be set in the file acq_parameters.py.
To run one simulation, set the parameters (with or without a mirror system) in the file simulate_one.py and execute it
python simulate_one.py
To run many simulations with varying number of irrelevant actions, set the corresponding parameters in the file simulate_many.py and execute it
python simulate_many.py
The figures bellow depict the desirability curves of the actions and the trial lengths with and without the mirror system for 25 irrelevant actions. The solid vertical line signifies the moment of the lesion, whereas the dashed vertical line is the time of recovery, i.e. the time when the "cat" found the new optimal strategy.
| With the mirror system | Without the mirror system |
|---|---|
 |
 |
 |
 |
 |
 |
 |
 |
The figure bellow shows the mean trials (along with the standard error) required for recovery after the lesion as a function of the number of irrelevant actions for an agent with the mirror system and without.
The main difference, in contrast to the paper, is that the agent without the mirror system cannot find easily the optimal strategy even if only 1 irrelevant actions is available. But the important point, that as the number of irrelevant actions increases the agent without the mirror system needs more trials for recovery, but the agent with the mirror system is not affected, is present.
Bonaiuto, J., & Arbib, M. A. (2010). Extending the mirror neuron system model, II: what did I just do? A new role for mirror neurons. Biological cybernetics, 102(4), 341-359.