event thumbnail image
Machine Learning over Text & Images - Autumn School

Semisupervised Learning Approaches

author: Tom Mitchell, School of Computer Science, Carnegie Mellon University
You might be experiencing some problems with Your Video player.
Slides
0:00 Semi-Supervised Learning over Text
1:24 Statistical learning methods
4:14 Semi-supervised Document classification
4:38 Document Classification: Bag of Words Approach
5:04 Twenty NewsGroups
6:42 What if we have labels for only some documents?
12:48 Nigam et al.
14:49 E-step, M-step
16:20 Using one labeled example per class
20:37 20 Newgroups - 1
24:18 20 Newgroups - 2
24:20 Downweight the influence of unlabeled examples by factor lambda
24:21 Why/When will this work?
35:39 EM for Semi-Supervised Doc Classification
35:41 Using Redundantly Predictive Features
37:25 Redundantly Predictive Features
40:20 Co-Training - 1
42:59 CoTraining Algorithm #1
44:02 - Redundantly Predictive Features - Part 2
45:23 CoTraining: Experimental Results
46:05 CoTraining setting
46:45 Co-Training Rote Learner
53:51 Expected Rote CoTraining error given m examples
54:50 - CoTraining setting - Part 2
55:40 - Co-Training Rote Learner - Part 2
59:11 What to Know
60:29 Further Reading - 1
60:44 Further Reading - 2

Lecture rating

People found this lecture:
Worth seeing
because it is:
 Valuable and informative
Well presented
Easily understandable
Acceptably recorded
You need to login to cast your vote.

Report a problem or upload files

If you have found a problem with this lecture or would like to send us extra material, articles, exercises, etc., please use our ticket system to describe your request and upload the data.
Enter your e-mail into the 'Cc' field, and we will keep you updated with your request's status.

Related content

Link this page

Would you like to put a link to this lecture on your homepage?
Go ahead! Copy the HTML snippet !

Reviews and comments:

Comment1 weiwei_kebi_marburg, January 5, 2008 at 6:17 p.m.:

Really a nice lecture!

Thanks a lot for uploading it.

Comment2 Atif Abdul-Rahman, February 12, 2008 at 11:02 p.m.:

hmm....very interesting for first timers in semi-supervised learning, but i was also expecting more on worse case scenario in both EM and Co-Training approaches he discussed. Wished he had more time to speak, still a great start....

Comment3 Tim Graettinger, April 29, 2008 at 3:54 a.m.:

The talk was very well done. The use of co-training seems artificial to me, however. By segregating the feature sets into 2 groups, no new information has been created. Why not just train a classifier using all of the features? It's not like the labels are only known for one classifier or the other. If that was true, there wouldn't be any way for classifier one to help classifier two.

Comment4 Petko, July 10, 2008 at 12:37 p.m.:

It was a very nice and understandable lecture.
I would like to know how one actually finds the groups of redundant predictive features. In theory there could be more than two groups, which means that you may extend the co-training in dimensionality. I cannot estimate although if that might help get a way better accuracy.
I also expected to learn more for the worst case scenarios..

Comment5 newton, March 16, 2009 at 10:21 a.m.:

The slides seem to be corrupt. It is not opening. Kindly address this issue.

Comment6 A Chess, January 3, 2010 at 4:55 a.m.:

Excellent lecture, Tom Mitchell rocks!

Write your own review or comment:

make sure you have javascript enabled or clear this field: