:일반화한다. 1. Dropout regularization 2.
Inverted Dropout Dropout의 변수로 Keepprob 라는 것을 두어, Keepprob(Ex. 0.8)라는 확률로 층마다 유닛을 유지할지를 결정한다. Dropout = np.random.rand(Actiovationlayer.shape[0], Actiovationlayer.shape[1]) # Step 1: initialize matrix Dropout = np.random.rand(..., ...)
Dropout = (Dropout < keep_prob).astype(int) # Step 2: convert entries of Dropout to 0 or 1 (using keep_prob as the threshold) Actiovationlayer = Actiovationlayer * Dropout # Step 3: shut down some neurons of Actiovationlayer Act...
#
Dropout
#
Frobenius
#
Norm
#
Overfitting
#
Regularization
#
Weight_decay
원문 링크 : Regularization
