Combining Active Learning and Semi-Supervised Learning Using Gaussian Fields and Harmonic Functions

DMCA

Combining Active Learning and Semi-Supervised Learning Using Gaussian Fields and Harmonic Functions (2003)

Cached

Download Links

by Xiaojin Zhu , John Lafferty , Zoubin Ghahramani

Venue:	ICML 2003 workshop on The Continuum from Labeled to Unlabeled Data in Machine Learning and Data Mining
Citations:	120 - 5 self

BibTeX

@INPROCEEDINGS{Zhu03combiningactive,
    author = {Xiaojin Zhu and John Lafferty and Zoubin Ghahramani},
    title = {Combining Active Learning and Semi-Supervised Learning Using Gaussian Fields and Harmonic Functions},
    booktitle = {ICML 2003 workshop on The Continuum from Labeled to Unlabeled Data in Machine Learning and Data Mining},
    year = {2003},
    pages = {58--65}
}

OpenURL

Abstract

Active and semi-supervised learning are important techniques when labeled data are scarce. We combine the two under a Gaussian random field model. Labeled and unlabeled data are represented as vertices in a weighted graph, with edge weights encoding the similarity between instances. The semi-supervised learning problem is then formulated in terms of a Gaussian random field on this graph, the mean of which is characterized in terms of harmonic functions. Active learning is performed on top of the semisupervised learning scheme by greedily selecting queries from the unlabeled data to minimize the estimated expected classification error (risk); in the case of Gaussian fields the risk is efficiently computed using matrix methods. We present experimental results on synthetic data, handwritten digit recognition, and text classification tasks. The active learning scheme requires a much smaller number of queries to achieve high accuracy compared with random query selection. 1.

Keyphrases

harmonic function active learning semi-supervised learning using gaussian field edge weight high accuracy gaussian field unlabeled data semi-supervised learning handwritten digit recognition text classification task present experimental result gaussian random field classification error semisupervised learning scheme matrix method synthetic data active learning scheme weighted graph gaussian random field model semi-supervised learning problem random query selection important technique

DMCA