1. 机器学习就是电脑系统自动根据每次运行经验来提升表现,降低损失,从而做出更好的预测,也就是自动从海量数据里提取到(或者说学习到)有用信息。Machine learning is all about learn INFORMATION from DATA, which could be anything from text and video to image and sound.
2. 预测(in supervised learning):
a.预测未来的事件:股票价格、医疗风险、行车事故风险
b.预测properties we don't yet know: 比如would I like this movie? 我喜欢什么样的电影?这张图片是关于什么?这句话是什么意思?类似像这样的问题很难hard code,首先它的属性范围可能很多,其次是我可能也不知道从什么属性着手去分析这个问题,这很难去specify,不好回答。但是,举例子就很容易,我知道我喜欢《电影a》、《电影b》...,于是根据这些经验,就能大概预测出来我是否会喜欢《电影c》。It‘s easier to express tasks in terms of examples of what you want (rather than how to solve them).
所以,the power of ML now is in terms of automating the process of finding the solution to a task that's illustrated by examples.
3. Reinforcement learning: learning to act, not just predict, goal to optimize the consequences of actions.