|
Title:
|
BATCH QUERY SELECTION IN ACTIVE LEARNING |
|
Author(s):
|
Piotr Juszczak |
|
ISBN:
|
978-972-8924-88-1 |
|
Editors:
|
Ajith P. Abraham |
|
Year:
|
2009 |
|
Edition:
|
Single |
|
Keywords:
|
active-learning, multiple query selection |
|
Type:
|
Full Paper |
|
First Page:
|
35 |
|
Last Page:
|
42 |
|
Language:
|
English |
|
Cover:
|
|
|
Full Contents:
|
click to dowload 
|
|
Paper Abstract:
|
In the active learning framework it is assumed that initially a small training set Xt and a large unlabelled set
Xu are given. The goal is to select the most informative object from Xu. The most informative object is the
one, that after revealing its true label by the expert, and adding it to the training set improves the knowledge
about the underlying problem the most, e.g. improves the most the performance of a classifier in a classification
problem. In some practical problems however, it is necessary to select at the same time more than a single
unlabelled object to be labelled by the expert. In this paper, we study pitfalls and merits of such selection. We
introduce active learning functions that are especially useful in the multiple query selection. The performance
of the proposed algorithms are compared with standard single query selection algorithms on toy problems and
the UCI repository data sets. |
|
|
|
|
|
|
