supervised learning is what we have done and it actually works very well but it relies on these large quantities of not just large data but requires a
bid lable data and that's a big constraint ,and where unsupervised learning is to develop techniques that can uncover or learn useful structure using very large quantities o unlabled data. Learning generative models
will be a way that we try to learn structure in our data using a lot o lable data but without requiring human annotation on that data.
参考
https://www.bilibili.com/video/BV1R5411a7eA?p=19
CS231n进阶!斯坦福李飞飞弟子密歇根大学老师《深度学习与计算机视觉》第19课