This note is the lecture note for 李宏毅(NTU) machine learning 2022

Lecture 1:Introduction of Deep Learning

Preparation

Pre1

  • 机器学习实际上是通过机器构造一个隐式的函数

  • two main function of machine learning

    • Regression: The function outputs a scalar
    • Classification
    • Structured Learning: create something with structure(image, document)

  • Label 学习中的真实值$y$（预测值为$\hat{y}$）

​ feature input所蕴含的某种信息

  • Does local minima truly cause the problem?

Pre2

  • 多个piecewise linear curve可以逼近任意曲线

    用sigmoid函数表达一个piecewise linear curve

​ image-20220915221127215

  • epoch与batch

​ image-20220915235644151

Class Material

week1

  • Supervised Learning, Self-supervised Learning, RL…

pytorch1

Lecture 2:What to do if my network fails to train

Preparation

Class Material

week2

  • 模型复杂度model complexity

    $\mathcal{H}$是模型参数可取值的集合，用其元素个数$\left | \mathcal{H} \right |$表示模型复杂度

    image-20220916164234379