Context Cells: Towards Lifelong Learning in Activity Recognition Systems
A robust activity and context-recognition system must be capable of operating over a long period of time, exploiting new sources of information as they become available and evolving in an autonomous manner, coping with user variability and changes in the number and type of available sensors. In particular, wearable and ambient nodes should be trained lifelong, as new context instances naturally arise, and the labeling of the instances should be carried out ideally with no user intervention. In this paper we show by means of an experiment and simulations that we can indeed achieve lifelong learning and automatic labeling by using Context Cells, an architecture capable of sensing, learning, classifying data and exchanging information.
Unable to display preview. Download preview PDF.
- 5.Harms, H., Amft, O., Roggen, D., Tröster, G.: Smash: A distributed sensing and processing garment for the classification of upper body postures. In: Third interational conference on body area networks (2008)Google Scholar
- 6.Bao, L., Intille, S.S.: Activity recognition from user-annotated acceleration data. In: Pervasive Computing: Proc. of the 2nd Int’l Conference, pp. 1–17 (2004)Google Scholar
- 8.SPINE: http://spine.tilab.com/ (Last seen May 15, 2009)
- 9.Lombriser, C., Roggen, D., Stäger, M., Tröster, G.: Titan: A tiny task network for dynamically reconfigurable heterogeneous sensor networks. In: 15. Fachtagung Kommunikation in Verteilten Systemen (KiVS), pp. 127–138 (2007)Google Scholar
- 13.Angluin, D., Laird, P.D.: Learning from noisy examples. Machine Learning 2(4), 343–370 (1988)Google Scholar