Yahoo Answers is shutting down on May 4th, 2021 (Eastern Time) and beginning April 20th, 2021 (Eastern Time) the Yahoo Answers website will be in read-only mode. There will be no changes to other Yahoo properties or services, or your Yahoo account. You can find more information about the Yahoo Answers shutdown and how to download your data on this help page.

? asked in 社會與文化語言 · 1 decade ago


Since these methods have the nonlinear function

such as a sigmoidal function as the unit function

and some intermediate layers, they can easily construct

a nonlinear mapping function within the NN

and have the generalization ability to respond correctly

to inputs it has not specifically been trained for.

It should be noted, however, that they are easy to fall

into a local minimum, because the back-propagation

algorithm is basically a gradient method. In addition,

the network scale will become very large for a

multi input-output system such as the control problem

of robot manipulators; in particular, the number

of units in the intermediate layers increases [17] and

hence the learning process is not implemented effectively.

A method [18] is recently proposed by adjusting

the unit functions and the connection weights,

or only the unit functions, instead of adjusting the

only connection weights as done in the conventional

method. It is shown in [19],[20] that this methodgives

a faster learning process and less number of unit functions

in the intermediate layers than the conventional

met hod.


2 Answers

  • 1 decade ago
    Favorite Answer



        因為這些方法有非線性作用譬如一個乙狀結腸的作用作為單位作用和一些中間層數, 他們能容易地修建一個非線性映射函數在NN 之內和有它不具體地被訓練了為的概念化能力正確地反應輸入。它應該是著名, 然而, 他們容易分成一個地方極小值, 因為傳播算法基本上是梯度方法。另外, 網路標度將變得非常大為一個多輸入- 輸出系統譬如機器人操作器的控制問題; 特別是, 單位的數量在中間層數增加[

    17 ] 並且因此學習進程有效地不被實施。方法[

    18 ] 由調整單位作用和連接重量, 或唯一單位作用最近提議, 而不是調整唯一的連接重量依照做在常規方法。它比常規遇見的煤斗被顯示[ 19], [20 ] 這methodgives 一個更加快速的學習進程和較少單位作用的數字在中間層數。


  • 因為這些方法安排非線性起作用. 例如一個乙狀結腸的作用作為部件功能并且一些中間層,他們可以容易地修建在NN之內的一個非線性映射函數并且有概念化能力恰當地反應對輸入它不明確地被訓練了為。應該注意,然而,他們是容易落入一個地方極小值,因為back-propagation算法基本上是梯度法。 另外,

    網絡等級將變得非常大為a多輸入-輸出系統例如控制問題機器人操作器; 特別是,數字在中間層增量[17的]單位和因此學習進程沒有有效地被實施。方法[18]通過調整最近提議部件功能和連接重量,或者仅部件功能,而不是調整仅連接重量如做在常規方法。 這methodgives的顯示[19], [20]一個更加快速的學習進程和較少單位的數量作用在中間層比常規遇見的煤斗。

Still have questions? Get your answers by asking now.