机器学习是让机器具有学习的能力,looking for a function from data
人工智慧是我们想要达成的目标,而机器学习是达成目标的手段,深度学习是机器学习中的一种方法。
机器学习三个步骤:
step1. 确定function
step2. 衡量function的好坏
step3. 选出最好的function
Learning Map
blue:scenario场景 不同的data使用不用的scenario
red :task任务 随着function,output的不同而不同,如regression、classification
green:method方法 同样的task可以用不同的model解决
task color: red(task任务 随着function,output的不同而不同,如regression、classification )
1. Regression: The output of the target function F is "scalar", eg. predict PM2.5 tomorrow
2. Classifiction: Binary Classification(output: Yes/No ; eg. Spam filtering), Multi-class Classification(output: Class 1,Class 2, ... , Class N ; eg. Document Classification)
3. Structured Learning - Beyongd Classifiction (输出为有结构性的东西)
scenario color: blue(scenario场景 不同的data使用不用的scenario)
1. Supervised Learning(缺点:hard to collect a large amount of labelled data) 监督学习
Classification - Deep Learning:Image Recognition, Playing Go, ...
Training Data: Input/Output pair of target function
Function(model)
output = label
2. Semi-supervised Learning 半监督学习
3. Transfer Learning 迁移学习
4. Unsupervised Learning--Machine Reading, Machine Drawing 无监督学习
5. Reinforcement(需要一个对手,一般为另一个机器)强化学习
Alpha Go is supervised learning + reinforcement learning
reinforce learning是没有办法做supervised learning时才做的(直接被pass掉?)
difference between supervised learning and reinforcement:
supervised: learning from teacher
reinforcement: learning from critics