NIPS*98 Workshop - Integrating Supervised and Unsupervised Learning

NIPS*98 Workshop
``Integrating Supervised and Unsupervised Learning''
Friday, December 4, 1998


Rich Caruana
Virginia de Sa
Michael Kearns
Andrew McCallum


This was a one day workshop held on Friday, Dec. 4, 1998. The format was a mix of short presentations and group discussions. There were two half-hour invited presentations to motivate the morning and afternoon sessions. Following were short (< 10 or 5 minute) presentations where invited attendees were able to lay one issue or method on the public table.


This workshop debated the relationship between supervised and unsupervised learning. The discussion ran the gamut from examining the view that supervised learning can be performed by unsupervised learning of the joint distribution between the inputs and targets, to discussion of how natural learning systems do supervised learning without explicit labels, to the presentation of practical methods of combining supervised and unsupervised learning by using unsupervised clustering or unlabelled data to augment a labelled corpus. The debate was fun because some attendees believe supervised learning has clear advantages, while others believe unsupervised learning is the only game worth playing in the long run.

More specifically, we discussed many of the following topics.

One goal for the workshop is to take existing methods that do both supervised and unsupervised learning and "plot" them along a few important dimensions. Our hope was to synthesize a description of current methods that makes the similarities and differences between methods more obvious, and which makes new opportunities more obvious.


Morning Session
7:30-8:15 Invited Talk: Michael Kearns (AT&T Labs)
Hear RealAudio of his talk.
8:15-9:15 Tradeoffs in Computing Joint Densities vs. Conditional Densities
Pedro Domingos (University of Washington)
Alexander Gray and Dennis Decoste (JPL)
Trevor Hastie and Dan Rubenstein (Stanford)
Tony Jebara and Alex Pentland (MIT)
Machiel Westerdijk and Dave Barber (University of Nijmegen)
9:15-10:30 Using Unlabelled Data to Help Supervised Learning
Kristin Bennet and Ayhan Demiriz (Rensselaer Polytechnic)
Zehra Cataltepe (Bell Labs)
Nathan Intrantor (Tel-Aviv University)
Nathalie Japkowicz (Princeton)
Andrew McCallum (Just Research & CMU)
Tom Mitchell (CMU)

Extracurricular Session

Afternoon Session
4:00-4:45 Invited Talk: Naftali Tishby (Hebrew University)
Hear RealAudio of his talk.
4:45-5:30 Theories of Unsupervised Learning and Missing Values
Leo Breiman (UC Berkeley)
Joachim Buhmann (University of Bonn)
Doug Fisher (Vanderbilt)
Nir Friedman (UC Berkeley) and Moises Goldszmidt (SRI)
Volker Tresp (Siemens)
5:30-5:55 Cognitive Theories
Virginia de Sa (UCSF)
Randall O'Reilly (University of Colorado)
5:55-6:20 Creating Fake Labels
Cyril Goutte, Jan Larsen, and Lars Hansen (Technical University of Denmark)
Chris Thornton (University of Sussex)
6:20-6:45 There is Unsupervised Learning in Supervised Learning
Juergen Schmidhuber (IDSIA)
Rich Caruana (Just Research & CMU)
6:45-7:00 Wrap-Up


The deadline for submitting abstracts was October 15th. If you are interested in presenting at the workshop but have not yet sent us an abstract, it is unlikely that we would be able to squeeze you into the presentation schedule. However, we encourage you to come to the workshop and contribute to the discussions. Please contact the workshop organizers at

Last updated: 9 November 1998