Abstract
A new approach to the construction and optimization of local models is proposed in the context of data streams, that is, unlimited sources of data where the utilization of all observations is impractical. Real-time revision of the learning set allows selective coverage of regions in state space which contribute most to reconstructing the underlying dynamical system.
- Received 13 May 2003
DOI:https://doi.org/10.1103/PhysRevLett.92.164101
©2004 American Physical Society