定义

Old

Field of study that gives computers the ability to learn without being explicitly programmed —— Arthur Samuel (1959)

Mordern

A computer program is said to learn from experience E with respect to some class of tasks T and performance measure P, if its performance at tasks in T, as measured by P, improves with experience E —— Tom Mitchell (1998)

算法分类

  1. 监督学习(Supervised Learning) 特点：”Right answers given” 即有训练集(training set)，包含输入和期望的输出
    • 回归 (Regression) Predict continous valued output
    • 分类 (Classification) Discreted valued output（判断是或不是）
  2. 非监督学习(Unsupervised Learning) 即没有训练集，譬如典型的根据相似度进行分类，但并不知道某一类是什么
  3. 其他还有增强学习(Reinforcement Learning), 推荐系统(Recommender Systems)