1.batch learning
batch of (email,spam)------>spam filter
batch of (patient,cancer)-------->cancer filter
batch of patient data-------->group of patients
batch learning:a very common protocol.
batch spam filter:predict with fixed g(固定的g)
2.online learning---which sequentially.
online learning不是固定的g(t)
PLA很容易就应用在online protocol;
reinforcement learning is often done online.因为强化学习也是sequentially。
online:hypothesis “improves” through receiving data instances sequentially.会不断地对g(t)进行修正
3.Active learning ----"learning by asking"
protocol等价于learning philosophy(学习哲学)
主动学习和之前的区别就在于,它会再回到前面的目标函数f上去“问”选定的xn对应的yn。
Active:improve hypothesis with fewer labels (hopefully) by asking questions.
多应用在labels标记很昂贵的情况下,比如珍贵药物标记。
总结: