Part Ⅰ

Part Ⅰ

1.1 Words List

• machine learning 機器學習

• algorithm 算法

• regression 迴歸

• classcational problem 分類問題

• benign && malignance 良性 & 惡性

• Machine Learning definition

  Arthur Samuel Machine Learning: Field of study that gives computers the ablity to lean without being explictly programmed.

  這是一個非正式的定義：在不直接針對問題進行編程的情況下，賦予計算機學習能力的一個研究領域。

  Tom Mitchell(1998) well_posed Learning problem: A computer program is said to learn from experience E with respect to some task T and some performance measure P if its performance on T as measured by P improves with experience E.
  學習問題應該這樣定義：對於一個計算機程序來說，給它一個任務T和一個性能測量方法P如果在經驗E的影響下P對T的測量得到了該進，那麼就說該程序從E中得到了學習。

1.2 Acquisition

• Four major subjects:

  1. Supervised Learning 監督學習

     Baidu’s definition of Supersived Learning:

     The process of adjusting the parameters of a classifer to adsired the performance using a set of known categories of samples,also known as Supervised Training or Teather Learning.

     利用一組已知類別的樣本調整的分類器的參數，使其達到所要求性能的過程，也稱爲監督訓練或有教師學習。

     For example:We have a group data about the prices and area of house,the prices as y axis and the area as x axis. So we can apply the data to get some relationships about x and y.These maybe are linear or norlinear and we can use these relationships to get the prices of the house when we know the area of the house.This remembers me of a course named as Calculation method in the last semester when I was a junior student.One of the experiments of the course called Least squares was a kind of way of supervised learning.

  2. Learning Theory 學習理論

     Emphasis on applied the learning theory to the practical problems.

  3. Unsupervised Learning 無監督學習

     Baidu’s definition of Unsupervised Learning:

     The various problems in pattern recognition are solved according to the training samples whose categories are unknown(not marked),which is recalled Unsupervised Learning.

     根據類別未知(沒有被標記)的訓練樣本解決模式識別中的各種問題，稱之爲無監督學習。

     Well-known clustering is an example of a typical unspervised learning.

  4. Reinforcement Learning 強化學習

     Agents learn in a “trial and error” way is called Reforcement Learning.By continuous learning,we tell agents which one is correct and which one is wrong.After training,the agents have the ablity judge right or wrong independently so that take right decidion in next step.