In deep neural networks, what primary role do "biases" serve alongside weight matrices?

Question

Accepted Answer

They shift the activation function horizontally, allowing the network to fit data that does not pass directly through the origin.

Answer

They enforce strict orthogonality between hidden layers.

Answer

They scale the variance of the input features to match a normal distribution.

Answer

They dynamically adjust the learning rate during stochastic gradient descent.

Crowdly