Machine Learning

     Recently, I am studying Maching Learning which is our course. My English is not good but this course use English all, and so I use English to record my studying notes. And our teacher is Dr.Deng Cai and reference book is Pattern Classfication. This is only my studing notes and I want to share to this blog.

 

     OK, first question, what is Machine Learning?

• Machine learning is the study of computer systems that improve their performance through experience.
• Learn existing and known structures and rules.
• Discover new findings and structures. (e.g Face recognition, Face recognition)

 For example:

                                       

                           dog                                                                                                       cat

Now I want to ask you the follow picture is what?

                                                               

This is a typical case of Machine Learning (or  Pattern Classification).

     Second, some terminology:

• Supervised learning vs. Unsupervised learning （监督学习和非监督学习，注：这里二者的主要区别在于监督学习给出了类别而非监督学习是要自己去聚合的即自己找出类别）
• Training data & test data （训练数据和测试数据）
• Supervised learning: Classification, Categorization, Decision making
• Unsupervised learning: Clustering, Matrix factorization, Topic modeling

     Third, how pattern system work:

• Domain-specific knowledge (e.g Acquisition, representation)
• Data acquisition (e.g camera, ultrasound, MRI,….)
• Preprocessing (e.g Image enhancement, segmentation)
• Representation (e.g Features: color, shape, texture,…)
• Decision making (e.g use of Statistical (geometric) pattern recognition or Artificial neural networks)
• Post-processing; use of context

     Finally, an example:Fish Classification:Salmon v. Sea Bass （注：例子是对传送带上的与进行分类，两种鱼）

     We first use the length as feature:Representation: Fish Length As Feature (注：这里中文解释下，就是比较两个鱼的长度，用来进行分类）

 

     We next use the lightness as feature:Fish Lightness As Feature（同上，用光泽度）

     Two-dimensional Feature Space

 

     Complex Decision Boundary

 

 

     Now we know something about Machine Learning and how to classifiy the test data according to training data, but one thing must be attention:

• Simple model à large training error, less test error
• Complex model à less training error, large test error 

     OK, that's all.