MA 490 Pattern Recognition

Using Hidden Markov Models

Winter Quarter 2009-10

shibberu@rose-hulman.edu

What is an HMM?

Hidden Markov Models (HMMs) are the powerful pattern recognition algorithms used in speech and optical character recognition software. Rose-Hulman's telephone operator is an HMM. HMMs are currently being used to:

- identify music by sound

- filter spam email

- analyze stock market data

- identify genes in DNA sequences

- model animal behavior

As more people gain familiarity with HMMs, the number and variety of applications is likely to increase. Collaborative projects between Biology and CS/Math Majors are especially encouraged. (Email me.) See past student projects below.

CS/Software Eng. Majors: An HMM is a probabilistic version of a finite state machine.

Why do HMMs work well in practice?

HMMs have a simple, yet rich structure which enables them to be specially tailored to a particular application. HMM algorithms can also be shown to be optimal (i.e. best in class).

What do I need to know to take this course?

Either MA 223 Statistics or MA 381 Probability provides sufficient background. All other concepts will be developed from scratch.

How will my grade be computed?

30% Homework

30% In Class Lessons/Quizzes

30% Project

10% Class Participation

Past Student Projects

Biology

Stochastic Context-Free Grammars and tRNA Folding
Michael Ewing, Mike Simon and Phil Smith

Robotics

Textured Image Segmentation Using Wavelet-Domain Hidden Markov Trees
Jay Groven and Alex Van Brunt