spiral recurrent neural network for online learning：网上学习的螺旋递归神经网络.pdfVIP

ESANN2007 proceedings - European Symposium on Artificial Neural Networks Bruges (Belgium), 25-27 April 2007, d-side publi., ISBN 2-930307-07-2. Spiral Recurrent Neural Network for Online Learning 1 2 1 Huaien Gao , Rudolf Sollacher , Hans-Peter Kriegel 1- University of Munich, Germany 2- Siemens AG, Corporate Technology, Germany Abstract. Autonomous, self* sensor networks require sensor nodes with a certain degree of “intelligence”. An elementary component of such an “intelligence” is the ability to learn online predicting sensor values. We consider recurrent neural network (RNN) models trained with an extended Kalman filter algorithm based on real time recurrent learning (RTRL) with teacher forcing. We compared the performance of conventional neural network architectures with that of spiral recurrent neural networks (Spiral RNN) - a novel RNN architecture combining a trainable hidden recurrent layer with the “echo state” property of echo state neural networks (ESN). We found that this novel RNN architecture shows more stable performance and faster convergence. 1 Introduction Sensor networks are going to be deployed in many control and diagnosis ap- plications where one expects solutions with low effort for installation, config- uration and operational maintenance. Those “plugplay” requirements imply autonomous behavior and forbid extensive parameter tuning. This also requires a certain degree of self-organization with little or no a priori knowledge of the environment. Such kind of “intelligence” can be achieved with self-learning sys- tems. One important component of such a self-learni