Hierarchical neural model: L3
Wen-Kuang Chou, D.Y.Y. Yun · 1991
It is observed that none of the currently popular learning algorithms are in-place learning algorithms. Instead of finding an in-place learning algorithm, which is considered impossible by the authors, a hierarchical neural model (L3) is proposed. L3 consists of a massively parallel architecture for recalling (MPAR) and a learning heuristics controller (LHC). Two operation modes of neural networks, recalling and learning, were realized by the two solid architectures (MPAR and LHC). Due to the separability of these two architectures, L3 has rechargeable capability. As a result, the function of L3 is very similar to the programmable logic array. The significance of L3 lies in the solid architecture of MPAR and LHC, and the rechargeable capability.>