pytorch（四）——笔记

目录

• 1.过拟合与欠拟合
• • 1.1 Scenario1:house price
  • 1.2 Scenario2: GPA
  • 1.3Model Capacity
• 2. 交叉验证
• • 2.1 Splitting
  • 2.2 test while train
  • 2.3 train-val-test
• 3.Regularization
• • 3.1 Occam's Razor
  • 3.2 Intuition
  • 3.3 L2-regularization
• 4. 动量与学习率衰减
• • 4.1 momentum：动量，也叫惯性
  • 4.2 momentum代码
  • 4.3 Learning rate tunning
  • 4.4 Learning rate decay
• 5. 什么是卷积
• • 5.1 What’s wrong with Linear
  • 5.2 Receptive Field
  • 5.3 Weight sharing
  • 5.4 Rethink Linear layer
• 6. 卷积神经网络
• • 6.1 Convolution
  • 6.2 Moving window
  • • 6.3 Several kernels
  • 6.4 Pyramid Architecture
  • 6.5 nn.Conv2d

1.过拟合与欠拟合 1.1 Scenario1:house price

1.2 Scenario2: GPA

1.3Model Capacity

2. 交叉验证 2.1 Splitting

2.2 test while train

2.3 train-val-test

3.Regularization 3.1 Occam’s Razor

右边的噪声太大了

3.2 Intuition

对比

3.3 L2-regularization

代码实现

4. 动量与学习率衰减 4.1 momentum：动量，也叫惯性

4.2 momentum代码

4.3 Learning rate tunning

4.4 Learning rate decay

5. 什么是卷积 5.1 What’s wrong with Linear

线性层

5.2 Receptive Field

5.3 Weight sharing

通过每一次感受小的视野来判断

局部相关性

5.4 Rethink Linear layer

6. 卷积神经网络 6.1 Convolution

6.2 Moving window

取决于观察视角

6.3 Several kernels

代表不同观察角度的信息

6.4 Pyramid Architecture

6.5 nn.Conv2d

代码实现