Anytime modeling: Compression and improvement of the approximation of SC based mappings

Nowadays practical solutions of engineering problems involve model-integrated computing. Due to their flexibility, robustness, and easy interpretability, the application of soft computing based models, may have an exceptional role. Despite of their advantages, the usage is still limited by their exponentially increasing computational complexity. Although, combining soft computing and anytime techniques it becomes possible to cope with the available, usually imperfect or even missing information, the dynamically changing, possibly insufficient amount of resources and reaction time. In this paper, possibilities offered by (Higher Order) Singular Value Decomposition ((HO)SVD) based anytime Soft Computational (SC) models are analyzed. With the help of SVD, the redundancy of fuzzy systems and neural nets can be removed and further, non-exact reduction can also be obtained. The method also offers a way for improving the model if later on we get into possession of new information or more resources. An algorithm is suggested, which finds the common minimal implementation space of the compressed original and the new approximation points, thus the two techniques, non-exact complexity reduction and improvement of the approximation accuracy, ensure that we can always cope with the temporarily available (finite) time/resources and find the balance between accuracy and complexity.