its really simple... in rough terms it breaks loops that your estimator might get itself in! Imagine you estimator is a dynamic system, and while its convergin

its really simple... in rough terms it breaks loops that your estimator might get itself in! Imagine you estimator is a dynamic system, and while its converging it get trapped in a cycle between two values. Adding a little noise can break this cycle.