李宏毅机器学习01introduction
-
机器学习就是自动找一个函数的过程,给定输入得到想要的输出
regression
classification
generation(产生文字、画图…) -
supervised learning
已经知道想要的输出是什么样子
给机器一些训练集(labeled data)<x,y>
机器根据loss function判断找的预测函数对不对,机器会自动找出loss最低的函数
-
reinforcement learning
不需要告诉机器怎么做,机器自己学习后得到一个reward,以 试错 的方式学习
-
函数寻找方法:gradient descent
-
explainable AI 可解释性AI
-
adversarial attack 增加噪声
network compression 模型压缩
the technique of optimising a deep learning model by reducing its number of parameters. In our paper, we do this by computing l2 norm of orcale ranking based on first order Taylor expansion for each layer followed by removal of parameters which are less than a given threshold and retraining the network. -
meta learning元学习
-
learn to learn 学习如何学习
-
life long learning终身学习
不断学习,越来越厉害