'
EECE515 Machine Learning, Spring 2007
Announcements

May 19

Reminder: The makeup class, Monday 7:00 pm, May 21.

No class on Thursday, May 31, due to my trip to MSRA, Beijing.
A makeup class will be scheduled in June.

Report for hwk #5 should be direclty emailed to Jiho Yoo
by midnight, May 31 .

June 5

We will have the makeup class on Wednesday, 3:00 pm, June 6.

No class on Thursday, June 7.
Lectures
Homework Assignments

Hwk 2
 Run your own hard/soft kmeans clustering algorithm on toy data sets.
 Download data sets
( Data 1 )
( Data 2 )
( Data 3 )
 Each data set contains two 2dimensional data points, x an y.
 Run your algorithms and describe what you found.
 Image segementation with hard/soft kmeans.
 Download Berkeley Segmentation data set
( Tiger )
( Airplane )
 Try your kmeans with at least 5 different values of k (e.g., k=2,3,5,7,13).
 Describe what you found.
 Due March 29 (Thu)

Hwk 3
 Do detailed derivation of EM algorithm for MoG,
in the case of arbitrary covariance matrices.
 Apply your MoG to the same tasks as homework #2.
 Due Apr 5 (Thu)

Hwk 4
 Derive the EM algorithm for maximum likelihood factor analysis.
 Computer exercise for EMFA.
 Implement your EMFA algorithm and apply it to
Iris data and
Wine data which are available from
UCI Repository .
 You need to investigate whether latent structure is really
useful (see TippingBishop's paper).
 Due April 24 (Tue)

Hwk 5
 Computer exercise for face recognition: FLD+MLP.
 Use ORL face dataset.
 ORL dataset Contains 400 face images
(10 images for each person and 40 people in total) of size 45 by 35.
 Implement FLD and use it as feature extraction.
 Implement MLP and use it as a classifier.
 You can play with different numbers of features (1020) using FLD.
 Try crossvalidation to select appropriate number of
hidden neurons.
 Due Midnight, May 31 (Thu).
 Directly send your report to Jiho Yoo (zentasis@postech.ac.kr)
and cccopy to me (seungjin@postech.ac.kr) as well.

Hwk 6
 Computer exercise for face recognition: KPCA+MLP
 First, derive the detailed algorithm for KPCA.
(have a look at this paper )
 Use ORL face dataset.
 ORL dataset Contains 400 face images
(10 images for each person and 40 people in total) of size 45 by 35.
 Implement KPCA and use it as feature extraction.
 Use your MLP you used in hwk #5.
 Compare KPCA results with PCA results.
 Due Midnight, June 25 (Mon).
 Directly send your report to Jiho Yoo (zentasis@postech.ac.kr)
and cccopy to me (seungjin@postech.ac.kr) as well.