Hebbian Learning Rule
A purely local neural network learning rule proposed by Donald Hebb in 1949. When two neurons are repeatedly activated at the same time, the strength of the connection between them increases, summarized as "neurons that fire together wire together." Mathematically, the weight change is proportional to the product of the activation values of the pre- and post-synaptic neurons, requiring neither a global error signal nor external supervision.
Timeline (last 90 days)
The mathematical form of the Hebbian learning rule is usually expressed as the weight change being proportional to the product of the activation values of the two neurons before and after.
Pure Hebbian rules face an instability problem: positive feedback may cause weights to grow without bound.
To improve the stability of the Hebbian rule, researchers proposed the Oja rule (introducing weight decay to achieve normalization) and the BCM rule (introducing a sliding modification threshold).
All Facts (3)
The mathematical form of the Hebbian learning rule is usually expressed as the weight change being proportional to the product of the activation values of the two neurons before and after.
50%UnverifiedPure Hebbian rules face an instability problem: positive feedback may cause weights to grow without bound.
50%UnverifiedTo improve the stability of the Hebbian rule, researchers proposed the Oja rule (introducing weight decay to achieve normalization) and the BCM rule (introducing a sliding modification threshold).
50%