Environment-adaptive learning: How clustering helps to obtain good training data

Debnath S, Baishya SS, Triebel R, Dutt V, Cremers D (2014)

Publication Type: Book chapter / Article in edited volumes

Publication year: 2014

Journal

Lecture Notes in Computer Science Springer Verlag

Publisher: Springer Verlag

Series: Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

Book Volume: 8736

Pages Range: 68-79

DOI: 10.1007/978-3-319-11206-0_8

Abstract

In this paper, we propose a method to combine unsupervised and semi-supervised learning (SSL) into a system that is able to adaptively learn objects in a given environment with very little user interaction. The main idea of our approach is that clustering methods can help to reduce the number of required label queries from user interaction, and at the same time provide the potential to select useful data to learn from. In contrast to standard methods, we train our classifier only on data from the actual environment and only if the clustering gives enough evidence that the data is relevant. We apply our method to the problem of object detection in indoor environments, for which we use a region-of-interest detector before learning. In experiments we show that our adaptive SSL method can outperform the standard non-adaptive supervised approach on an indoor office data set.

Involved external institutions

Technische Universität München (TUM) DE Germany (DE) Indian Institute of Technology IIT Mandi IN India (IN)

How to cite

APA:

Debnath, S., Baishya, S.S., Triebel, R., Dutt, V., & Cremers, D. (2014). Environment-adaptive learning: How clustering helps to obtain good training data. In (pp. 68-79). Springer Verlag.

MLA:

Debnath, Shoubhik, et al. "Environment-adaptive learning: How clustering helps to obtain good training data." Springer Verlag, 2014. 68-79.

BibTeX: Download