Zum Lernverhalten von Backpropagation-Netzen auf der Basis stochastischer Rechentechnik (On the Learning Process of Back-Propagation Nets on the Basis of Stochastic Computation)
Abstract
The dissertation entails a detailed examination of the adaptive or learning process of a back-propagation net, used for the building of massive parallel neural networks, and based on the level of mathematical induction, stochastic computation, and software simulation. Firstly are the introductory bases of the back-propagation new discussed. Parameters and an adaptive process are then selected, followed by limitations and countermeasures. The results of the examination and software simulation demonstrate the simultaneously occurring work and learning processes do not seem to exert any negative influence on the convergence progress. The more precise and quantitative scalability of the entire process is evaluated in the study's final chapter.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jun 01, 2000
- Accession Number
- ADA407931
Entities
People
- Liyun Zhu