Line 16: | Line 16: | ||
==Relevant Rhea Pages== | ==Relevant Rhea Pages== | ||
*[[Lecture_6_-_Discriminant_Functions_Old_Kiwi|Spring 2008-Lecture 6, student lecture notes]] | *[[Lecture_6_-_Discriminant_Functions_Old_Kiwi|Spring 2008-Lecture 6, student lecture notes]] | ||
+ | *[[Bayes_Classification:_Experiments_and_Notes_OldKiwi|The effect of adding correlated features]] (some inspiration for the [[hw1_ECE662_S12|first homework]]?) | ||
*[[Bayes_Rate_Fallacy:_Bayes_Rules_under_Severe_Class_Imbalance|A student page about the effect of severe class imabalance]] (a topic mentioned in class today) | *[[Bayes_Rate_Fallacy:_Bayes_Rules_under_Severe_Class_Imbalance|A student page about the effect of severe class imabalance]] (a topic mentioned in class today) | ||
Revision as of 13:09, 2 February 2012
Lecture 8 Blog, ECE662 Spring 2012, Prof. Boutin
Thursday February second, 2012 (Week 4)
Today we continued our study of the separating hypersurface in the case where, for all class $ i=1,\ldots,c $,
$ \Sigma_i=\sigma^2 {\mathbb I}. $
We noted the co-dimension two of the intersections of the segments of hyperplanes forming the decision boundary. We also drew a connection with a shape analysis tool called the "skeleton" of a shape.
We then slightly generalized our study to the case where the standard deviation matrix is the same for all classes $ i=1,\ldots,c $. We then noticed the presence of the Mahalanobis distance in the discriminant function, and derived the relationship between the Mahalanobis distance and the Euclidean distance through a simple change of coordinates. It was pointed out by Mark that this change of coordinates is called "whitening".
We also spent a lot of time discussing the first homework.
Relevant Rhea Pages
- Spring 2008-Lecture 6, student lecture notes
- The effect of adding correlated features (some inspiration for the first homework?)
- A student page about the effect of severe class imabalance (a topic mentioned in class today)
Previous: Lecture 7
Next: Lecture 9
Comments
Please write your comments and questions below.
- Write a comment here
- Write another comment here.