Lecture 2 Blog, ECE662 Spring 2012, Prof. Boutin
Thursday January 12, 2012 (Week 1)
We began the lecture by summarizing the statistical pattern recognition paradigm with a simple flow chart. We then replaced this flow chart by a more complex one representing the pattern recognition paradigm from an engineering perspective, where the data acquisition and pre-processing are be included in the decision process.
Switching gears, we went back to our toy example to illustrate the concepts of "decision boundaries" and "discriminant functions". Our example only included one feature (hair length), so the corresponding decision boundaries in this case are points in a one-dimensional space. We then added another feature (shoulder width) to illustrate how the decision boundaries then become lines in a two-dimensional space.
As a warning to NOT use decision boundaries/discriminant functions for all problems, we considered the problem of determining the gender of a person from its first name. While the name can be written as a vector (e.g., each letter written as the integer representing its order in the alphabet), it would be silly to attack such a problem by trying to find a discriminant function. It is much better to look up all the names (e.g., on the US social security website), and to write down the most likely gender for each name in a look-up-table.
Previous: Lecture 1
Next: Lecture 3
Comments
Please write your comments and questions below.
- Write a comment here
- Write another comment here.
