Neural Net Evolution. A neural net typically starts out with random values for bias and weights; hence, it produces essentially random predictions when presented with its first case. What is the key ingredient by which the net evolves to produce a more accurate prediction?