Learning

Don Hebb's initial hypothesis was that learning only happened when two neurons were simultaneously active. There was experimental neural evidence of this in the 50s. He also considered the forgetting rule: if one neuron was active and another inactive than the strength between them was decreased. In 1950 there was no experimental neural evidence of this, but there is now.

CANT assumes that there is Hebbian Learning.

CANT assumes there is Anti-Hebbian Learning, though it is not committed on the pre but not post or post but not pre question.

These are the only types of learning that have significant experimental neural evidence. They are the only types of learning that CANT does.


CANT only learns from patterns of activation. If a neuron is never activated it can't be brought into a CA.

To avoid this problem, neurons can spontaneous activate. If a neuron has not been activated for a long time, it may spontaneously fire. This can enable it to be brought into CAs, and thus unused neurons are not wasted.


There is a vast range of possible Hebbian and Anti-Hebbian learning rules. These experiments suggest that it is best to add lots of strength when a neuron has little connection strength, and little when it is already well connected.

Taking strength from other axons also seems to be useful.

This is just evidential, not conclusive.