Learning Data Representations or Similarity Measures:
The UTL Challenge
Teleconference presentation, by Isabelle Guyon
Thursday January 20, 2011, 8 am PT, 11 am ET, 5 pm CET
Participation instructions http://www.afia-france.org/tiki-index.php?page=GroupeDeLecture110120IG (please review them in advance)
Is it possible to LEARN from unlabeled data Representations of Similarity Measures for use in supervised learning tasks? The Unsupervised and Transfer Learning challenge offers an opportunity to explore this problem of fundamental and practical interest.
Labeling data is not only expensive, it is tedious. When it comes to your own personal data it is also something you do not want to outsource. To help us tagging fast our personal pictures, videos, and documents, we need systems that can learn with very few training examples. The question is whether we can exploit similar data (labeled with different types of labels or completely unlabeled) to improve data preprocessing.
This presentation will outline the setup of the challenge and review the state of the art in unsupervised and transfer learning. Potential challenge participants are invited to attend and ask questions.
Prizes: $6000 + free registrations + travel awards
Dissemination: Workshops at ICML and IJCNN; proceedings in JMLR W&CP.
Deadline phase 1 (unsupervised learning): February 28, 2011
Deadline phase 2 (transfer learning): April 15, 2011
Challenge website: http://clopinet.com/ul