This is the code for the arxiv preprint Learning to acquire novel cognitive tasks with evolution, plasticity and meta-meta-learning.
We evolve plastic (i.e. self-modifying) neural networks, that can automatically acquire novel cognitive (meta-learning) tasks, not seen during training, from stimuli and rewards alone.
Note: This code is currently being revised.
The code is in the form of Jupyter notebooks that can be run on Google Colab.
It is strongly recommended to consult the Simple
notebook, which contains a simplified version of the code that should be easier to read through, while still producing the same results.
The other notebook contains the full code that was actually used to run the experiments.