The weights of synthetic neural networks should be initialized to small random numbers. It’s because that is an expectation of the stochastic optimization algorithm used to coach the mannequin, referred to as stochastic gradient descent….