EE509
Fuzzy
Logic and applications
Homework #6
Due: Tuesday, 11/21/05
1.
Implement a fuzzy pattern classifier for digits 0-9.
Requirements:
-
The digits are patterns on a 29x29 bitmap. The digits are
normalized to 27x27 size. 
-
Follow the example from the class notes to convert
the binary bitmap in to a fuzzy feature set.
- Use the supplied training data.
-
Use genetic algorithms to produce a fuzzy classifier which is using
only fuzzy operators AND and NOT (similar to the class example, but not
using OR).
- For each classifier map the features on a
color-coded (and numbered) graph and supply this graph in your report.
(For example,
if a square is included into the classification formula directly, color
it blue. If a square is included into the classification formula with
NOT, color it red. If a square is not included into the classification
formula, color it white.)
-
The final classifier program must take a file in
the format of the training set as the input, show the scores for all
digits from 0 to 9, and show the recognized digit.
Materials and hints:
- The training data and Matlab functions to access them are supplied here.
- Run
readme.m to look at all training patterns.
- Depending
on how you have implemented the fuzzify() function in the fuzzy
toolbox, you may want to reuse it for computing fuzzy features.
- Train a
single digit classifier at a time.
- A
suggestion for the fitness function: a classifier will have maximum
fitness when it recognizes all correct patterns in the training set
with a score of 1, and scores all other patterns in the training set
with a score of 0. (Example: all paterns for digit 3 are recognized as
3 with a score of 1.0, while patterns for digits 0-2,4-9 are scored as
0.0)
- Follow
the in-class discussion on the specifics of implementation.
Grading:
60 pts. The classifier correctly
classifies all patterns from the training set.
10 pts. Report satisfying all requirements listed in the
syllabus and this assignment.
3 pts max.
For correct classification of each of the 10 test patterns to be
presented to the classifier.
- If a pattern cannot be recognized by anyone's classifier, it will be replaced
- If a pattern can be recognized only by a single person's classifier, that person scores +3pts. everybody else scores +1pts.
- If a pattern can be recognized by two classifiers only, those persons score +3pts, everybody else scores +0.5pts.
- If a
pattern can be recognized by three or more classifiers, those persons
score +3pts, everybody else scores 0pts.
Last Updated:
08/25/03
Contact the instructor at: esazonov@clarkson.edu