Location via proxy:   [ UP ]  
[Report a bug]   [Manage cookies]                
Scribd
Upload
14 views

View Detection Algorithm

This document discusses active learning with multiple views. It presents three recent developments that reduce the need for labeled training data in multi-view learning: 1) Co-Testing, a general-purpose multi-view active learner that outperforms existing approaches; 2) Co-EMT, a multi-view learner that obtains robust behavior over different tasks using active and semi-supervised learning; 3) Adaptive View Validation, a meta-learner that predicts if multi-view learning is appropriate for new tasks based on past experiences.

Uploaded by

Srinivas
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
14 views

View Detection Algorithm

This document discusses active learning with multiple views. It presents three recent developments that reduce the need for labeled training data in multi-view learning: 1) Co-Testing, a general-purpose multi-view active learner that outperforms existing approaches; 2) Co-EMT, a multi-view learner that obtains robust behavior over different tasks using active and semi-supervised learning; 3) Adaptive View Validation, a meta-learner that predicts if multi-view learning is appropriate for new tasks based on past experiences.

Uploaded by

Srinivas
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 1

Active Learning with Multiple Views

view detection algorithm that automatically partitions Jones, R., Ghani, R., Mitchell, T., & Riloff, E. (2003).
a domain’s features in views that are adequate for Active learning for information extraction with mul-
multi-view learning. Such an algorithm would remove tiple view feature sets. Proceedings of the ECML-2003
the last stumbling block against the wide applicability Workshop on Adaptive Text Extraction and Mining.
of multi-view learning (i.e., the requirement that the
Knoblock, C. et al. (2001). The Ariadne approach
user provides the views to be used). Second, in order
to Web-based information integration. International
to reduce the computational costs of active learning
Journal of Cooperative Information Sources, 10,
(re-training after each query is CPU-intensive), one
145-169.
must consider look-ahead’ strategies that detect and
propose (near) optimal sets of queries. Finally, Adap- Muslea, I. (2002). Active learning with multiple views
tive View Validation has the limitation that it must be [doctoral thesis]. Los Angeles: Department of Computer
trained separately for each application domain (e.g., Science, University of Southern California.
once for wrapper induction, once for text classification,
etc.). A major improvement would be a domain-inde- Muslea, I., Minton, S., & Knoblock, C. (2000). Selec-
pendent view validation algorithm that, once trained tive sampling with redundant views. Proceedings of
on a mixture of tasks from various domains, can be the National Conference on Artificial Intelligence
applied to any new learning task, independently of its (AAAI-2000).
application domain. Muslea, I., Minton, S., & Knoblock, C. (2001). Hierar-
chical wrapper induction for semi-structured sources.
Journal of Autonomous Agents & Multi-Agent Systems,
CONCLUSION 4, 93-114.

In this article, we focus on three recent developments Muslea, I., Minton, S., & Knoblock, C. (2002a). Ac-
that, in the context of multi-view learning, reduce the tive + semi-supervised learning = robust multi-view
need for labeled training data. learning. Proceedings of the International Conference
on Machine Learning (ICML-2002).
• Co-Testing: A general-purpose, multi-view active Muslea, I., Minton, S., & Knoblock, C. (2002b). Adap-
learner that outperforms existing approaches on tive view validation: A first step towards automatic view
a variety of real-world domains. detection. Proceedings of the International Conference
• Co-EMT: A multi-view learner that obtains a on Machine Learning (ICML-2002).
robust behavior over a wide spectrum of learning
tasks by interleaving active and semi-supervised Muslea, I., Minton, S., & Knoblock, C. (2003). Active
multi-view learning. learning with strong and weak views: A case study on
• Adaptive View Validation: A meta-learner that wrapper induction. Proceedings of the International
uses past experiences to predict whether multi- Joint Conference on Artificial Intelligence (IJCAI-
view learning is appropriate for a new unseen 2003).
learning task. Nigam, K., & Ghani, R. (2000). Analyzing the effec-
tiveness and applicability of co-training. Proceedings
of the Conference on Information and Knowledge
REFERENCES Management (CIKM-2000).

Blum, A., & Mitchell, T. (1998). Combining labeled Nigam, K., McCallum, A., Thrun, S., & Mitchell, T.
and unlabeled data with co-training. Proceedings of (2000). Text classification from labeled and unlabeled
the Conference on Computational Learning Theory documents using EM. Machine Learning, 39(2-3),
(COLT-1998). 103-134.

Collins, M., & Singer, Y. (1999). Unsupervised models Pierce, D., & Cardie, C. (2001). Limitations of co-training
for named entity classification. Empirical Methods in for natural language learning from large datasets. Empiri-
Natural Language Processing & Very Large Corpora cal Methods in Natural Language Processing, 1-10.
(pp. 100-110).

0

You might also like