A study of neural network architecture for weak non-linear modeling
- Resource Type
- Conference
- Authors
- Mizukami, Y.; Wakasa, Y.; Tanaka, K.
- Source
- SICE 2004 Annual Conference SICE 2004 SICE 2004 Annual Conference. 1:548-551 vol. 1 2004
- Subject
- Robotics and Control Systems
Signal Processing and Analysis
Aerospace
Neural networks
Noise robustness
Linearity
Mathematical model
Reproducibility of results
Radial basis function networks
Ear
Nonlinear equations
- Language
This paper studies a property of neural network architecture for non-linear modeling. This method was proposed in our previous work and has three improvements; 1) the design of a sigmoidal function with localized derivative, 2) a deterministic scheme for weight initialization, and 3) an updating rule for weight parameters. We discuss its robustness against noise based on simulation results.