Lukas' Notes

neural-networks

Definition

Hidden Artificial Neuron

A hidden artificial neuron is an artificial neuron in a multi-layer perceptron that sits between the input and output layers. Its activations are not directly constrained by the training data — only the final output is supervised.

This is the source of the credit assignment problem: a hidden neuron has no teacher. Its target output must be inferred indirectly from how it contributes to the final prediction.