原ppt下载:pan.baidu.com/s/1nv54p9R,密码:3mty
需深入实践并理解的重要概念:
Deep Learning:
- SoftMax Fuction(输出层归一化函数,与sigmoid相似的激活函数,用于解决分类问题(分类大于2时;sigmoid解决二分类问题))
1)
2)每个neuron的softmax输出:,其中:
DNN(Deep Neural Networks):
- MSE(Means Square Error,均方误差) / CE(Cross Entropy,交叉熵)
- Use to minimum total costs for softmax layer. CE is better.
- MSE minimum:
- CE minimum:
- Mini-batch & batch_size(decides how many examples in a mini-batch) & epoch(周期)
- batch:样本训练中,将完整数据分为等量的多个batch(批次),每次输入一个batch而不是完整样本进行训练
- epoch:周期被定义为向前和向后传播中所有batch的单次训练迭代
- mini-batch has better performance than original gradient descent
- Vanishing Gradient Problem(梯度消失问题)
- ReLU(Rectified Linear Unit,线性纠正单元)
- As an activative function, used when the number of layers is quite large.
- 对于大于0的所有输入来说,它都有一个不变的导数值;常数导数值有助于网络训练进行得更快,常用于多层隐藏层
- Special cases of MaxOut:
- Learnable activation function
- Adaptive learning rate(学习率:每次迭代中cost function中最小化的量。简单来说,我们下降到cost function的最小值的速率是学习率)
- Use a large rate first, then change to a small one
- Momentum(动量原理)
- Use the optimizer Adam(Advanced Adagrad Momentum)
- Overfitting Problem(过拟合问题)
- Use early stopping
- Weight Decay(训练时用p%的dropout,测试时对权值做(1-p%)的调整后再获得输出)
- Dropout(训练的过程舍弃神经元)
- Will change structure of networks while training. better than MaxOut
CNN(Convolutional Neural Networks):
- Image recognization suits to use CNN because of 3 important properties:
1) Patterns are much smaller than the whole image
2) The same patterns appear in different regions
3) Subsampling pixels does not change the object
- filter & channel
- stride(step)
- zero-padding
- max-pooling
- flattern
- less parameters because of sharing weights
本站文章如无特殊说明,均为本站原创,如若转载,请注明出处:读《Deep Learning Tutorial》(台湾大学 李宏毅 深度学习教学ppt)后杂记 - Python技术站