随机梯度下降算法事实上是梯度下降算法的一个扩展，在深度学习中的代价函数

Stochastic gradient descent algorithm is in fact an extension of the gradient descent algorithm, the cost function in depth learning can be decomposed into the sum of the cost function for each sample. So with the growth in the size of the training set, step gradient calculation will consume a very long time. The core is stochastic gradient descent gradient is desired. Expect a small scale can be used to approximate the sample

The random gradient descent algorithm is in fact an extension of the gradient descent algorithm, and in deep learning the cost function can be broken down into the sum of each sample cost function. So as the training set grows in size, calculating a step gradient can also take a long time. At the heart of the random gradient descent is that the gradient is expected. Expect to use small sample approximations

In fact, the random gradient descent algorithm is an extension of the gradient descent algorithm. In depth learning, the cost function can be decomposed into the sum of each sample cost function. Therefore, it will take a long time to calculate one-step gradient with the increase of training set size. The core of random gradient descent is that gradient is expectation. It is expected that a small sample size approximation can be used<br>