Machine Learning in Wireless Sensor Networks, Tracking and Behavior Recognition (Localization, Tracking, Dimension Reduction, Manifold Regularization)

Machine Learning Approaches in Wireless Sensor Networks (Localization, Tracking, Dimension Reduction, Manifold, etc.)

►Distributed Intelligence◄
Localization and Tracking in Wireless Sensor Networks

Overview

Wireless sensor networks have attracted great interests in Artificial Intelligence and Machine Learning communities. Many tasks that range from robotics to context-aware computing can now be realized with the help of distributed wireless sensor networks. However, many of these applications assume availability of sufficient calibrated data, which are expensive to obtain. We have been developing Machine Learning techniques for relieving human labor and avoiding the use of costly calibration hardware. Thus, the deployment and learning process could be more intelligent and automatic. General techniques include unsupervised, semi-supervised and supervised learning.
LeMan presents a manifold regularization approach to calibration-effort reduction for tracking a mobile node in a wireless sensor network. Many previous approaches to the location-estimation problem assume the availability of calibrated data. However, to obtain such data requires great effort. We compute a subspace mapping function between the signal space and the physical space by using a small amount of labeled data and a large amount of unlabeled data. This mapping function can be used online to determine the location of mobile nodes in a sensor network based on the signals received. We use Crossbow MICA2 to setup the network and USB camera array to obtain the ground truth. Experimental results show that we can achieve a higher accuracy with much less calibration effort as compared to several previous systems.
Co-Localization addresses the problem of recovering the locations of both mobile devices and access points from radio signals by exploiting both labeled and unlabeled data from both mobile devices and access points. We first propose a solution using Latent Semantic Indexing to construct the relative locations of the mobile devices and access points when their absolute locations are unknown. We then propose a semi-supervised learning algorithm based on manifold to obtain the absolute locations of the devices. Both solutions are finally combined together in terms of graph Laplacian. Extensive experiments are conducted in wireless local-area networks, wireless sensor networks and radio frequency identification networks. The experimental results show that we can achieve high accuracy with much less calibration effort as compared to several previous systems.

Tracking Experiments

Test-bed for Wireless Sensor Networks

Test-bed for WiFi

Movies

Publications

Online Co-Localization in Indoor Wireless Networks by Dimension Reduction. [Paper] [Slides] [Bibtex]
Jeffrey J. Pan, Qiang Yang, Jialin Pan. In Proceedings of the 22nd National Conference on Artificial Intelligence (AAAI-07) Regular paper.
Adaptive Localization in a Dynamic WiFi Environment Through Multi-view Learning. [Paper] [Slides] [Bibtex]
Jialin Pan, James T. Kwok, Qiang Yang, Jeffrey J. Pan. In Proceedings of the 22nd National Conference on Artificial Intelligence (AAAI-07) Regular paper.
Domain-Constrained Semi-Supervised Mining of Tracking Models in Sensor Networks. [Paper] [Slides] [Bibtex]
Rong Pan, Junhui Zhao, Wenchen Zheng, Jeffrey Junfeng Pan, Dou Shen, Jialin Pan, and Qiang Yang. ACM SIGKDD. San Jose, USA. August, 2007.
Co-Localization from Labeled and Unlabeled Data Using Graph Laplacian (in Wireless and Sensor Networks). [Paper][Slides][Bibtex]
Jeffrey J. Pan, Qiang Yang. In Proceedings of the Twentieth International Joint Conference on Artificial Intelligence (IJCAI-07). Hyderabad, India. Jan, 2007. Regular paper.
A Manifold Regularization Approach to Calibration Reduction for Sensor-Network Based Tracking. [Paper] [Slides] [Bibtex]
Jeffrey J. Pan, Qiang Yang, Hong Chang, Dit-Yan Yeung. In Proceedings of the Twenty-First National Conference on Artificial Intelligence (AAAI-06) Regular paper.
Multidimensional vector regression for accurate and low-cost location estimation in pervasive computing. [Paper] [Slides] [Bibtex]
Jeffrey J. Pan, James T. Kwok, Qiang Yang, Yiqiang Chen. IEEE Transactions on Knowledge and Data Engineering (TKDE). Sept. 2006. Regular paper.
Accurate and low-cost location estimation using kernels. [Paper] [Slides] [Bibtex]
Jeffrey J. Pan, James T. Kwok, Qiang Yang, Yiqiang Chen. In Proceedings of the Nineteenth International Joint Conference on Artificial Intelligence (IJCAI-05), July 2005. Regular paper.

Software

LeMan : Location Estimation using Manifold Regularization. The Matlab source code would be coming soon.

Resources

Distributed Sensor Network Localization and Tracking, Robotics, Vision and Sensor Networks Group, David Moore, MIT
Calibration and Tracking in Sensor Networks, MIT
MoteTrack, Konrad Lorincz, Harvard University
Intel Research Seattle
Intel Place Lab
RSE Lab (Dieter Fox)
Manifold Localization (Neal Patwari)
Localization Survey (Moustafa Youssef)
On Manifold Regularization, Mikhail Belkin, University of Chicago
TinyOS

Thank You

This page was created on January 9th, 2006 and was updated on July 24th, 2006.