Tutorial Title: Meta-learning
Prof. Christophe Giraud-Carrier
Brigham Young University
Department of Computer Science
This tutorial is about meta-learning. The primary goal of meta-learning is the understanding of the interaction between the mechanism of learning and the concrete contexts in which that mechanism is applicable. Meta-learning differs from base-learning in the scope of the level of adaptation. Whereas learning at the base-level focuses on accumulating experience on a specific learning task (e.g., credit rating, medical diagnosis, mine-rock discrimination, fraud detection, etc.), learning at the meta-level is concerned with accumulating experience on the performance of multiple applications of a learning system.
In order to “cross the chiasm” and allow machine learning and data mining algorithms to be more widely used outside research labs, our community must be able to design robust systems that offer support to practitioners. Current machine learning/data mining tools are only as powerful/useful as their users. One main aim of current research in the community is to develop meta-learning assistants, able to deal with the increasing number of models and techniques, and give advice dynamically on such issues as model selection and method combination.
Over the past few years, much work has been done in the area, scattered across journals and conferences. There is added value in bringing interested researchers and practitioners together around a tutorial that describes the state-of-the-art and highlights avenues of future work.
With its focus on applications of machine learning, ICMLA provides a natural setting for this tutorial. It is hoped that academics will be encouraged to pursue research in this important area, that practitioners will be encouraged to join the effort by sharing requirements, and that software developers will see an opportunity to enrich their tools by integrating meta-learning assistants in their systems.
The following material can be covered in a tutorial of about 2 hours. A full set of slides will be made available to participants, together with an extensive bibliography. I will address the following topics:
1. A brief history of past efforts in understanding learning mechanisms
2. What is meta-learning and why does it matter?
Model combination vs. model selection
No Free Lunch
3. Meta-learning for model selection
From domain to model class to algorithm
Statistical and information-theoretic meta-features
5. The road ahead
Algorithm characterization and clustering
Bio Sketch of Dr. Christophe Giraud-Carrier:
Dr Giraud-Carrier is an Associate Professor in the Department of Computer Science at Brigham Young University. Prior to coming to BYU in 2004, he was Senior Manager at ELCA, a Swiss IT services company, where his responsibilities included the capitalization of Data Mining expertise, responses to tenders and the management of various projects for companies, local governments and NGOs. Prior to ELCA, Dr Giraud-Carrier was Senior Lecturer in the Department of Computer Science at the University of Bristol, where he founded and led the Machine Learning Research Group. Christophe received the B.S., M.S., and Ph.D. in Computer Science at Brigham Young University in 1991, 1993, and 1994, respectively.
Dr Giraud-Carrier is the director of BYU's Data Mining Lab. From 1998 to 2001, he led the ESPRIT METAL Project, Europe's first long-term research project on metalearning. He has organized workshops on metalearning at both ICML, ECML and IJCNN since 1999. He is a co-author of the first book dedicated to metalearning: Metalearning: Applications to Data Mining by Springer, 2008