Bias node in BP

Hello

By BP, what do you mean please?

Are you talking about back-propagation ? Yes you are, I didn't see your tag

If so, the bias is not related to back-propagation, but to Mc Culloch & Pitts model.

Any network model that use Mc Culloch & Pitts model must use a bias, as it is described in any good explanation.

The bias permit to shift the activation (or transfert) function, along x axis.

The common way to implement it is to add a constant input (1 or -1 usually) in input vector of each node and make its weight (i.e the bias) evolve like any other weight.

An analogy in 2D : the bias is like c in the line equation a.x+b.y+c=0