专业英语:mini-batch的作用
What is mini-batch used for? Please use your own words to answer the question.
Mini-batch是在原数据集中被随机选取的小量的m个训练输入.
Mini batch is a small amount of training inputs randomly selected in the original data set.
是用来通过进行时间和精度的trade-off进行 随机选取小样本,对原始庞大的训练输入的梯度下降方向进行一个较好的估计.
It is used as a trade off between the time and precision to estimate the gradient descent direction of the huge original training input.
梯度下降算法的原理是沿着每次的当前位置的导数方向走一小步,一直走到最小值点.
The principle of the gradient descent is to take a small step along the derivative direction of the current position each time until it reaches the minimum value point.This requires high cost derivative calculation of “current position” each time. And because the datasets involved in Neural Network are often huge, such calculation decreases the trainning efficiency. Therefore, Stochastic Gradient Descent comes in together with “mini-batch” as its training inputs to reduce the cost of training time.
这就要求每次都需要我们对“当前位置”的导数进行计算.在机器学习中我们涉及到的数据集一般规模都较大,这样会使我们的学习效率变得低,时间开销较大.所以,在随机梯度下降算法中,利用mini-batch作为输入以减小训练的时间话费开销.
虽然利用minibatch进行的估算存在统计波动,但实际关心的是在某个方向上移动来使cost-C最小.这意味着我们无需做梯度的精度计算.
Although there are statistical fluctuations in the estimation using minibatch, the actual concern of our algorithm is to move in a certain direction to minimize the “cost-C”. This means that there is no need to be too strict with high precision.
总结来说,使用mini-batch可以使我们的模型训练和不断迭代的过程中节约很多时间,速度快,提高时效性.
In conclusion, using mini batch can save a lot of time in the process of model training and continuous epochs
