赞
踩
最常见的connection method:Fully Connet Feedforward Network
一个Network就是一个function
layer1的neuron会与layer2的每一个neuron相连接,因此称为Fully Connet Network;因为整个过程由layer1传到layer2,由后往前传,因此称为Feedforward Network。
Deep = Many hidden layers
通过矩阵形式计算:
Using parallel computing techniques to speed up matrix operation,写成矩阵运算的好处在于可以使用GPU加速,比用CPU运算更快。
You need to decide the network structure to let a good function in your function set.
Q&A:
1、需要多少层,以及每层有多少神经,需要根据试验和误差,以及经验来决定
2、the structure can be automatically determined. E.g. Evolutionary Artificial Neural Networks
3、we can design the network structure. E.g. Convolutional Neural Network(CNN)
Cross Entropy 越小越好
For all training data,计算total loss
最小化total loss,利用gradient descent。
Backpropagation:
Ref:http://speech.ee.ntu.edu.tw/~tlkagk/courses/MLDS_2015_2/Lecture/DNN%20backprop.ecm.mp4/index.html
为什么要使用deep learning,deeper is Better?
参考:
https://www.bilibili.com/video/BV1Ht411g7Ef
Copyright © 2003-2013 www.wpsshop.cn 版权所有,并保留所有权利。