什么是机器学习

机器学习的任务

An Example
Concepts:
- Model
- Feature
- Weight
- Bias
- Label
- Error Surface
- Learning Rate
- Hyperparameter
2.
(Loss 越大,参数越糟糕)
3.

Summary:

- Linear Model
Improvement
Concept:
- Model bias
- Activation Function
- (Hard) Sigmoid
- ReLU
- Epoch
- Update

Constant + (Sigmoid→)Hard Sigmoid → Piecewise Function → Continuous Curve
Sigmoid 调整


New Model

矩阵表示:
1.
2.
3.
Batch and Momentum
More Changes

- Add Layers
此即 神经网络、 深度学习
Deep = Many hidden layers
隐藏层 = 线性变换 + 激活函数
Overfitting(过拟合)

为什么要把学习变深,而不是变“胖”,只加 ReLU 之类的数量,而不加层数? —— 后续内容
Summary
