| Research Abstract |
As a new methodology of intelligent system, we have proposed an pattern information based active learning method (PBALM), which has a non-exact processing nature without complex computational formulas. The PBALM is formed on the basis of fuzzy concepts, striving for human-like information processing. Typical fuzzy rules. On the other hand, the PBALM features are ambiguous pattern processings. This simulates human brain nature in the modeling of processing information in pattern-like images rather than utilizing numerical interpretations, when humans acquire knowledge from complex targets.
The PBALM has plural processing units called IDS's, which are tasked with extracting useful information from a system subject to modeling. In order to realize an PBALM system which is capable of performing complex, large scale modeling with an excellent fault tolerant architecture, dedicated hardware for IDS is indispensable. For the purpose of developing such an intelligent hardware system, we tentatively designed an IDS hardware unit using an evaluation board available on the market, and verified how effective it would be.
Because we confirmed the effectiveness of dedicated IDS hardware with the evaluation board, we developed a new IDS processing unit called 'HIDS'. We designed a printed circuit board on which an IDS controller which ran at 100MHz, were mounted. In performance test results, in spite of the parallelization of only four HIDS boards, the performance of HIDS was far superior in the processing time in comparison to that of the software-based PBALM. By using HIDS units that we developed in this project, the PBALM system would be applied to modeling and control applications under demanding realtime conditions.
|
