2002音訊處理與辨識.pptVIP

2002 音訊處理與辨識 * Machine Learning forData Clustering and Pattern Recognition J.-S. Roger Jang (張智星) CS Dept., Tsing Hua Univ., Taiwan /jang jang@ 2002 音訊處理與辨識 * * Outline Introduction to pattern recognition (PR) Datasets Approaches to PR Distance functions Data reduction Data clustering Discussions * * References “Pattern Recognition and Image Analysis”, by E. Gose, R. Johnsonbaugh, and S. Jost, Prentice Hall, 1996 “ Algorithms for Clustering Data”, by A. K. Jain and R. C. Dubes, Prentice Hall, 1988 “Pattern Classification: A Unified View of Statistical and Neural Approaches”, by J. Schurmann, John Wiley, 1996 * * Pattern Recognition (PR) Also known as: Data/pattern classification 圖形辨識 or 樣式辨認 Goal: Automatic classification of objects based on measurable quantities 根據已知的資料，由外在的特性來推測出內在的本質 Example: 視其所以，觀其所由，察其所安，人焉廋哉？人焉廋哉？ Applications of PR * * Data Sets There are numerous datasets for testing machine learning algorithms for PR: UCI Machine Learning Repository Datasets from NEC Research Lab Face recognition dataset Many many more… Introduction of some of the datasets in UCI Machine Learning Repository Example of PR An application example of PR with the use of histogram analysis * * Classification Methods Distance modeling K-nearest-neighbor classifiers (KNNC) Decision boundary modeling Linear classifiers (LC) Linear discriminant function classifiers (LDFC) PDF modeling Quadratic classifiers (QC) Gaussian-mixture-model classifiers (GMMC) Possibility modeling Fuzzy classifiers (FC) * * Other Classifiers There are numerous different classifiers SVM (support vector machine) Neural networks CART (classification and regression tree) ID4.5 Random forests HMM (hidden Markov model) AdaBoost Maximum entropy Many more… * * Distance/Similarity Functions Most classifiers require the calculation of distance or similarity. So distance/similarity functions are essential in PR. More about distance/similarity functions * * Data Reduction Purpose: Reduce