premise
Neural networks with appropriate weights and bias, and bias the weights to adjust to fit the training data process called learning. Neural Network Learning divided into the following four steps.
Step 1 (mini-batch)
selecting a portion of the random data from the training data.
Step 2 (calculated gradient)
is calculated on the weight loss function of gradient parameters of each right.
Step 3 (update parameter)
the weight parameter in the gradient direction minor updating.
Step (repeat) 4
Repeat Step 1, Step 2, Step 3.