A MULTILEVEL NEUROMOLECULAR ARCHITECTURE THAT USES THE EXTRADIMENSIONAL BYPASS PRINCIPLE TO FACILITATE EVOLUTIONARY LEARNING

被引:24
作者
CHEN, JC [1 ]
CONRAD, M [1 ]
机构
[1] WAYNE STATE UNIV,DEPT COMP SCI,DETROIT,MI 48202
来源
PHYSICA D | 1994年 / 75卷 / 1-3期
基金
美国国家科学基金会;
关键词
D O I
10.1016/0167-2789(94)90295-X
中图分类号
O29 [应用数学];
学科分类号
070104 ;
摘要
A multilevel neuronal architecture integrating memory and internal dynamics is presented. The architecture has two levels of pattern processing neurons, called enzymatic neurons, orchestrated by two levels of memory access neurons, called reference neurons. The neural activity of the enzymatic neurons is controlled by information processing on the membrane cytoskeleton, modeled with cellular automats. The reference neuron selects a repertoire of enzymatic neurons to perform coherent functions. Evolution occurs at three levels: at the level of readout enzymes that respond to cytoskeletal signals, at the level of proteins that control signal flow in the cytoskeleton, and at the level of reference neurons that orchestrate the repertoire of enzymatic neurons. The system was tested in a maze-like environment in which the organism must learn to use patterns of barriers in its ''field of vision'' to find a target. The integrated system effectively employs synergies among the different levels of processing and learning to acquire pathfinding capabilities in a complicated environment. It also exhibits substantially greater problem solving power when the dimensionality of the system is increased by adding more types of components and more weak interactions.
引用
收藏
页码:417 / 437
页数:21
相关论文
共 35 条