Abstract
An adaptive on-line algorithm extending the learning of learning idea is proposed and theoretically motivated. Relying only on gradient flow information it can be applied to learning continuous functions or distributions, even when no explicit loss function is given and the Hessian is not available. The framework is applied for unsupervised and supervised learning. Its efficiency is demonstrated for drifting and switching non-stationary blind separation tasks of acoustic signals. Furthermore applications to classification (US postal service data set) and time-series prediction in changing environments are presented.
Original language | English |
---|---|
Pages (from-to) | 743-760 |
Number of pages | 18 |
Journal | Neural Networks |
Volume | 15 |
Issue number | 4-6 |
DOIs | |
Publication status | Published - 2002 Jun |
Bibliographical note
Funding Information:We thank the participants of the 1997 on-line learning workshop at the Newton Institute in Cambridge for interesting discussions. K.R.M., M.K. and A.Z. gratefully acknowledge partial financial support from DFG under contracts JA 379/91, MU 987/11 and the BMBF BCI project and the EU in the Neurocolt 2 and the BLISS project (IST-1999-14190).
Keywords
- Adaptive learning rate
- Blind source separation
- ICA
- On-line learning
- Stochastic gradient descent
- Supervised learning
- Unsupervised learning
ASJC Scopus subject areas
- Cognitive Neuroscience
- Artificial Intelligence