Why multifactor?
- Take a simple MLP. Let be the layer activation. is the input, is the second layer (first hidden layer). These are vectors, indexed like .
- Then or . is the nonlinear activation function (ReLU, sigmoid, etc.)
- In standard STDP the learning rule follows or if layer number is
- (but of course nobody thinks there 'numbers' on the 'layers' of the brain -- this is just referring to pre and post synaptic).
- In an artificial neural network, (Intuitively: the weight change is proportional to the error propagated from higher layers times the input activity) where where is the derivative of the nonlinear activation function, evaluated at a given activation.
-
-
-
-
-
-
|