Involves modifying the performance function, which is normally chosen to be the sum of squares of the network errors on the training set, by adding some fraction of the squares of the network weights
If a system is being simulated without softening, then close encounters between particles can occur, and sometimes 3- or more-body interactions can result in the formation of a closed binary or multiple star system To accurately integrate these tight orbits takes time-steps many orders of magnitude smaller than is typical for non-interacting stars Multiple star systems form an integral part of the evolution of globular clusters, and thus cannot be ignored Without special treatment the integration of tight binary star systems can soak up the majority of your CPU time and slow the simulation immensely [1] Regularization is the process of speeding up the simulation of these orbits by solving them analytically as perturbed two-body systems