Random initialization refers to the practice of using random numbers to initialize the weights of a machine learning model.

Random initialization is one way of performing symmetry breaking, which is the act of preventing all of the weights in the machine learning model from being the same.