7.Analysis of neural networks: a random matrix approach 4.Regularization Regularization is a main method to avoid overfitting,by introducing regularization term into loss model: w*=argmin L(y,f(w,x))+XR(w) 个 regularization term where A is the regularization parameter,and (w)=w N=15 6 06 7. Analysis of neural networks: a random matrix approach 4.Regularization Regularization is a main method to avoid overfitting, by introducing regularization term into loss model: ↑ regularization term where is the regularization parameter, and