01-01-课程介绍
- Prerequisite
- Orientation
- Assignment Schedule
- Machine Learning
- Deep Learning
- Different types of Functions
- Supervised Learning(Lecture 1-5)
- Generative Adversarial Network(Lecture 6)
- Self-supervised Learning(Lecture 7)
- Anomaly Detection(Lecture 8)
- Explainable AI(Lecture 9)
- Model Attack(Lecture 10)
- Domain Adaptation(Lecture 11)
- Reinforcement Learning(Lecture 12)
- Network Compression(Lecture 13)
- Life-long Learning(Lecture 14)
- Meta Learning(Lecture 15)
Prerequisite
Orientation
Assignment Schedule
Machine Learning
Machine Learning ≈ Looking for Function
机器学习:让计算机寻找一个复杂函数式。
Deep Learning
深度学习:让计算机通过神经网络寻找一个复杂函数式。
Different types of Functions
Supervised Learning(Lecture 1-5)
Generative Adversarial Network(Lecture 6)
Supervised Learning 需要成对的(样本,标签)数据集,而 Generative Adversarial Network 只需要样本,和标签,不需要两者之间的关联关系,GAN可以自己找到这种关联关系。
Self-supervised Learning(Lecture 7)
Supervised Learning的缺点是需要对大量样本进行标记,成本很高,而且,为每一个任务收集样本也很麻烦,所以产生了Self-supervised Learning。
Self-supervised Learning 不需要预先对样本进行标注,节省了很大成本。
现在开发一个application很简单,因为与硬件交互的事情都有操作系统帮忙处理,Pre-trained Model(Foundation Model) 可以类比为操作系统,Downstream Tasks 相当于application。
Anomaly Detection(Lecture 8)
Explainable AI(Lecture 9)
Model Attack(Lecture 10)
Domain Adaptation(Lecture 11)
Reinforcement Learning(Lecture 12)
Network Compression(Lecture 13)
Life-long Learning(Lecture 14)
Meta Learning(Lecture 15)
learn to learn