| 摘要 |
PURPOSE: A neural network algorithm implementation method is provided to extract optimal common terms and to minimize an addition network based on the common terms by using a neural network optimization algorithm, the Boltzman machines, when an addition network is implemented in a DA(Distributed Arithmetic) structure for an inner product. CONSTITUTION: The method comprises several steps. A graph is generated for expressing different coefficient patterns and the number of common terms based on common terms data of each bit for given coefficients(S111). It is determined whether common terms data exists on the graph(S112). If it exists, optimized data is extracted from the common data by using the Boltzman machines, and an addition network is constructed based on the extracted data(S113, S114). Then, if the addition network is constructed, the data on the bit included in grouping the common terms is eliminated for checking whether other common terms data exists(S115). The steps(S111-S115) are repeated, and then if the common terms data does not exist, the addition network is constructed by using the remaining bit data(S116).
|
