Learning from Point Sets with Observational Bias - Robotics Institute Carnegie Mellon University

Learning from Point Sets with Observational Bias

Liang Xiong and Jeff Schneider
Conference Paper, Proceedings of 30th Conference on Uncertainty in Artificial Intelligence (UAI '14), pp. 898 - 906, July, 2014

Abstract

Many objects can be represented as sets of multi-dimensional points. A common approach to learning from these point sets is to assume that each set is an i.i.d. sample from an unknown underlying distribution, and then estimate the similarities between these distributions. In realistic situations, however, the point sets are often subject to sampling biases due to variable or inconsistent observation actions. These biases can fundamentally change the observed distributions of points and distort the results of learning. In this paper we propose the use of conditional divergences to correct these distortions and learn from biased point sets effectively. Our empirical study shows that the proposed method can successfully correct the biases and achieve satisfactory learning performance.

BibTeX

@conference{Xiong-2014-17171,
author = {Liang Xiong and Jeff Schneider},
title = {Learning from Point Sets with Observational Bias},
booktitle = {Proceedings of 30th Conference on Uncertainty in Artificial Intelligence (UAI '14)},
year = {2014},
month = {July},
pages = {898 - 906},
}