Related Experiment Videos
High-order MS_CMAC neural network
1Department of Civil Engineering, National Chiao Tung University, Hsinchu, Taiwan 300, R.O.C.
IEEE Transactions on Neural Networks
|February 6, 2008
Summary
A new high-order macro structure cerebellar model articulation controller (HMS_CMAC) uses quadratic splines for smoother nonlinear system modeling. This approach improves generalization and reduces training data needs for complex problems.
Area of Science:
- Artificial Intelligence
- Machine Learning
- Neural Networks
Background:
- Traditional Cerebellar Model Articulation Controllers (CMAC) face computational complexity in multidimensional problems.
- Existing macro structure CMAC (MS_CMAC) with a trapezium scheme offers nonlinear system modeling but lacks smooth interpolation.
- Parameter tuning for the trapezium scheme requires extensive cross-validation.
Purpose of the Study:
- To develop a high-order MS_CMAC (HMS_CMAC) that overcomes the limitations of the trapezium scheme.
- To enable smooth interpolation and improve generalization in nonlinear system modeling.
- To reduce the computational complexity and training data requirements for CMACs.
Main Methods:
- A quadratic splines scheme was developed to replace the trapezium scheme in MS_CMAC.
- The quadratic splines scheme transforms stepwise weight contents into smooth weight contents for continuous outputs.
- The proposed HMS_CMAC utilizes a tree structure, decomposing multidimensional problems into 1-D subproblems.
Main Results:
- The HMS_CMAC demonstrates acceptable generalization capabilities in continuous function-mapping problems.
- The use of nonoverlapping association in training instances significantly reduces the number of required training samples.
- Only a single learning cycle is needed during the training stage with nonoverlapping association.
Conclusions:
- The HMS_CMAC effectively models nonlinear systems with smooth outputs, outperforming previous methods.
- The quadratic splines scheme enhances the performance and efficiency of MS_CMAC.
- HMS_CMAC offers a computationally efficient and data-efficient approach for complex function approximation.
