evolve_and_merge
#hebbian #phd
2024-02-12 07:13
ABCD, again. Just the same as meta-learning-through-hebbian-plasticity-KG2BPTBB The point in this one was to reduce the number of rules while also increasing generalizability. Simply superior.
It’s cool how they are very indirectly training the network. The parameters aren’t really the weights, but the plasticity parameters. Very neat.