Institutional Repository
Technical University of Crete
Technical University of Crete
EN
|
EL
| URI: | http://purl.tuc.gr/dl/dias/69D8E21E-7A00-4447-BA05-9EC5B1C9983E | ||
| Year | 1993 | ||
| Type of Item | Peer-Reviewed Journal Publication | ||
| License |
|
||
| Bibliographic Citation | M. P. Thint, P. P. Wang and A. Dollas, "Nonparametric graded data processing with back-error propagation networks", Inf. Sci., vol. 67, no. 1-2, pp. 167-188, Jan. 1993. doi:10.1016/0020-0255(93)90089-5 https://doi.org/10.1016/0020-0255(93)90089-5 | ||
| Appears in Collections |
We present some computational characteristics of back-error propagation (BEP) networks in processing graded patterns that are otherwise indistinguishable in binary (or bipolar) representations. We address the problem of mapping l-of-m unit gradients with interunit activation profile d to n classes, where maximum noise amplitude of ϵ is permitted within gradient classes. Relations between these parameters and the training period measured in epochs (T) are discussed. Extensions of basic concepts are used to extract embedded feature information in constrained 2D grey-scale patterns and to group and classify distorted pattern clusters whose intraset distances are sometimes greater than interset metrics. Results have been applied to simulation studies in the domain of robotic tactile pattern recognition.
Copyright © DIAS 2013
