Unique representations of dynamical systems produced by recurrent neural networks
This paper considers learning a dynamical system (DS) by a recurrent neural network (RNN). We propose an affine neural dynamical system (A-NDS) as a DS that an RNN actually produces on the output space to approximate a target DS. We present a unique parametric representation of A-NDSs using RNNs and affine sections with the aim of constructing effective learning algorithms.
Unable to display preview. Download preview PDF.
- 1.Baldi, P., Gradient descent learning algorithm overview: A general dynamical systems perspective, IEEE Transactions on Neural Networks 6 (1995), 182–195.Google Scholar
- 2.Hertz, J., Krogh, A., and Palmer, R. G., Introduction to the Theory of Neural Computation. Addison Wesley, 1991.Google Scholar
- 3.Hirsh, M. W., and Smale, S., Differential Equations, Dynamical Systems and Linear Algebra. Academic Press, 1974.Google Scholar
- 4.Hornik, K., Stinchcombe, M., and White, H., Multilayer feedforward networks are universal approximators, Neural Networks 2 (1989), 359–366.Google Scholar
- 5.Kimura, M., and Nakano, R., Learning dynamical systems from trajectories by continuous time recurrent neural networks, Proceedings of 1995 IEEE International Conference on Neural Networks 6 (1995), 2992–2997.Google Scholar
- 6.Kimura, M., and Nakano, R., Learning dynamical systems produced by recurrent neural networks, Proceedings of 1996 International Conference on Artificial Neural Networks (1996), 133–138.Google Scholar
- 7.Sussmann, H. J., Uniqueness of the weights for minimal feedforward nets with a given input-output map, Neural Networks 5 (1992), 589–593.Google Scholar
- 8.Tsung, F-S., and Cottrell, G. W., Phase-space learning, Advances in Neural Information Processing Systems 7 (1995), 481–488.Google Scholar