代價函式與學習規則（Cost Function and Learning Rule）

本節簡要說明了代價函式是什麼，以及它如何影響神經網路的學習規則。

This section briefly explains what the costfunction is and how it affects the learning rule of the neural network.

該函式也被稱為損失函式或目標函式。

It is also called the loss function andobjective function.

代價函式是一個與優化理論相聯絡的數學概念。

The cost function is a rather mathematicalconcept that is associated with the optimization theory.

你不必瞭解這個概念。

You don’t have to know it.

然而，如果你想要更好地理解神經網路的學習規則，那麼最好是能夠深刻理解它。

However, it is good to know if you want tobetter understand the learning rule of the neural network.

這並不是一個很難理解的概念。

It is not a difficult concept to follow.

代價函式與神經網路的監督學習有關。

The cost function is related to supervisedlearning of the neural network.

第二章闡述了神經網路的監督學習就是通過調整權值來減小訓練資料誤差的過程。

Chapter 2 addressed that supervisedlearning of the neural network is a process of adjusting the weights to reducethe error of the training data.

在這種情況下，代價函式是神經網路誤差的度量。

In this context, the measure of the neuralnetwork’s error is the cost function.

神經網路的誤差越大，則代價函式的值也越大。

The greater the error of the neuralnetwork, the higher the value of the cost function is.

神經網路的監督學習主要有兩種代價函式型別。

There are two primary types of costfunctions for the neural network’s supervised learning.

圖3-9 訓練輸出與正確輸出之間的差異越大，則誤差越大The greaterthe difference between the output and the correct output, the larger the error

顯然，代價函式值與誤差成正比。

It is clearly noticeable that the costfunction value is proportional to the error.

這種關係非常直觀，不需要進一步解釋。

This relationship is so intuitive that nofurther explanation is necessary.

神經網路的大多數早期研究都採用這種代價函式來推導學習規則。

Most early studies of the neural networkemployed this cost function to derive learning rules.

前一章的增量規則是從該函式匯出，而且反向傳播演算法也是如此。

Not only was the delta rule of the previouschapter derived from this function, but the back-propagation algorithm was aswell.

迴歸問題仍然使用這個代價函式。

Regression problems still use this costfunction.

——本文譯自Phil Kim所著的《Matlab Deep Learning》

更多精彩文章請關注微訊號：