1. Introduction

12th, Sept.

1.1. Contents of this lecture

  • Concepts
  • Math formulation
  • System construction
  • Theory and method
  • Typical algorithm
  • Newest problem

1.2. Concepts

  • observation -> decision
  • Pattern: a physical arrangement of elements
  • Recognition: an awareness that something perceived has been perceived before
  • Sample, class, features. Classify samples according to their features
  • System: object/pattern -> sensor -> feature extractor -> classifier -> decision/action
  • Methods:
    • Knowledge-based: AI, expert systems; syntax PR or structural PR
    • Data-based: statistical PR; ANN, SVM (This lecture)
    • Hybrid methods
  • Classifier, decision boundary

1.3. Eg: coin classification

  • No observation, decided between two classes
    • According to prior probabilities, choose bigger one
    • Minimize error rate
  • Have weight information
    • According to posterior probabilities
    • How to calculate posterior probabilities? Bias equation (below)
    • Risk of the decision

equation

1.4. Eg: How to do speech recognition

  • Consider all possible text, extract feature, calculate posterior probabilities and do classification.
  • Corpus
  • MFCC feature for speech