COLT '92 Proceedings of the fifth annual workshop on Computational learning theory
State-of-the-Art Review: A User's Guide to the Brave New World of Designing Simulation Experiments
INFORMS Journal on Computing
Toward the training of feed-forward neural networks with the D-optimum input sequence
IEEE Transactions on Neural Networks
| Hi-index | 0.00 |
D-optimal and other model-based experimental designs, like the Query by Committee criterion, are often criticised because of their dependency to the statistical model and the lack of the explicit allocation of the identified irregularities of the assumed model. Furthermore, D-optimal experimental designs tend to weight the boundary area of the experimental space significantly. In extreme cases the boundary points of the experimental space are the experimental candidates. In order to defuse this critique a new online Bayesian design for the nonlinear dynamic system identification is introduced, which serves the flexibility and which is concurrently more resistant against the bias caused by the model. An advantage of this approach is also the small modification of the D-optimal design to increase the independency of the D-optimal design to the chosen model approach. This contribution includes the presentation of different methods for the Design of Experiments for nonlinear dynamic system identification. Therefore, model-free and model-based based experimental designs are introduced and the new online Bayesian design for nonlinear dynamic system identification is represented. The presented design plans are appropriate to identify the dynamics of combustion engines, can be used to identify the parameters of both analytical and simulation models and are evaluated by simulations.