在BBC一檔關於數學的節目裏，數學家兼主持人談到了他對Pattern的癡迷。在他看來，數學就是對Pattern尋找和研究的過程。以下筆記將要涉及的當然不是他所說的數學也不是他所說的廣義的Pattern，而是曾經在大學裏因爲上課打盹或者逃課而沒能學完(一點都沒能)的模式識別。從一開始就想要把它學好學精，But God knows what I was afraid of? 沒有基石無以成大業。從頭來過！
所有筆記內容(以問答方式)均來自於學習《Pattern Recognition 4th Edition》(主要部分)以及我的拙見(very very 次要部分)。以下是該書的原版出版社鏈接：http://www.elsevierdirect.com/product.jsp?isbn=9781597492720
What is pattern recognition?
Pattern recognition is the scientific discipline whose goal is the classification of objects into a number of categories or classes.
In what field the pattern recognition will be applicable?
Machine vision, Character recognition, Computer-aided diagnosis, Speech recognition, and Data mining and knowledge discovery in data bases.
What is feature, feature vector, l-dimensional feature space, training patterns?
Feature is the measurements used for classification. In more general case, l features constitute a feature vector.
Feature vectors are sometimes treated as random variables as the measurements resulting from different patterns exhibit a random variation.
All feature vectors constitute an l-dimensional feature space. The classification is to separate the space with decision surfaces.
The patterns (feature vectors) whose true class is known (called labeled) and which are used for the design of the classifier are known as training patterns (training feature vectors).
What the basic stages to design a classification system are?
1: Feature Generation Stage
2: Feature Selection Stage
3: Classifier Design Stage
Optimality Criteria, Decision Surfaces
4: System Evaluation Stage
Classification Error Rate
What is supervised, unsupervised, and semi-supervised pattern recognition?
Supervised pattern recognition: A set of training data are available. The classifier is designed by exploiting the known information.
Unsupervised pattern recognition (Clustering): No available training data of known class labels. A set of feature vectors are given and the goal is to find the underlying similarities and cluster similar vectors together.
The major issue is to define the similarity between two vectors and choose an appropriate measure for it. Another issue is to choose an algorithmic scheme to cluster the vectors on basis of the adopted similarity measure.
Semi-supervised pattern recognition: Share the same goal with supervised pattern recognition. Labeled and unlabeled training data both exists. Recovering the related information from unlabeled data is useful to improve the system design. In other words, labeled data can be constraints in clustering task and provide knowledge that clustering algorithm has to respect.
