Pattern Processing and Control with a Virtual NeuromolecularComputer Architecture

9409780康拉德开发了一个多层次神经分子计算体系结构，为进化学习提供了丰富的平台。该体系结构包括一个由神经元状模块组成的网络，其内部动力学由细胞自动机建模。动力学反映了被认为在真实神经元中起作用的分子过程，特别是与第二信使信号和细胞骨架-膜相互作用有关的过程。目标是通过进化搜索算法创建专用动态模式处理器库，然后使用记忆操作算法从库中选择能够执行相干模式识别/神经控制任务的处理器组合。按照目前的实施，该系统由两层细胞骨架神经元和两层记忆通路神经元(称为参考神经元)组成，这些神经元被分成功能上可比较的子集的集合。进化学习可以通过负责信号在空间和时间上整合的细胞骨架结构的变化，或者通过代表读入或读出蛋白质的元件的位置的变化，在神经元内水平上发生。协调神经元处理器指令集的记忆操作算法也使用进化搜索程序，也很适合在联想模式下操作。综合系统有效地利用不同水平的处理和学习之间的协同作用，以在迷宫般的环境中获得寻路能力。近期的研究目标包括对系统的性能进行全面的实验表征，拓宽神经元的动态能力，研究系统的结构与自适应表面的“进化友好性”之间的关系，结合进化学习利用联想记忆能力，以及在各种应用领域研究系统。长期目标包括实现开放式进化学习，将具有有用模式处理能力的进化神经元移植到特殊的硅硬件上，将该系统用作新兴分子电子技术的体系结构范例，以及将该系统用作更清楚地了解神经元内机制在大脑功能中的作用的工具。***

9409780 Conrad A multilevel neuromolecular computing architecture has been developed that provides a rich platform for evolutionary learning. The architecture comprises a network of neuron-like modules with internal dynamics modeled by cellular automata. The dynamics reflect molecular processes believed to be operative in real neurons, in particular processes connected with second messenger signals and cytoskeleton-membrane interactions. The objective is to create a repertoire of special purpose dynamic pattern processors through and evolutionary search algorithm and then to use memory manipulation algorithms to select combinations of processors from the repertoire that are capable of performing coherent pattern recognition/neurocontrol tasks. The system, as presently implemented, consists of two layers of cytoskeletal neurons and two layers of memory access neurons (called reference neurons) divided into a collection of functionally comparable sublets. Evolutionary learning can occur at the intraneuronal level through variations in the cytoskeletal structures responsible for the integration of signals in space and time or through variations in the location of elements that represent readin or readout proteins. The memory manipulation algorithms that orchestrate the repertoire of neuronal processors also use evolutionary search procedures, and are well suited for operating in an associative mode as well. The integrated system effectively employs synergies among the different levels of processing and learning to acquire pathfinding capabilities in a maze-like environment. Immediate objectives of the research include full experimental characterization of the performance capabilities of the system, widening the dynamic capabilities of neurons, studying the relation between the structure of the system and the "evolution friendliness" of the adaptive surface, utilization of the associative memory capability in combination with evolutionary learning, and study of the system in a variety of a pplication domains. Long term goals include the achievement of open ended evolutionary learning, porting evolved neurons with useful pattern processing capabilities to special silicon hardware, use of the system as an architectural paradigm for emerging molecular electronic technologies, and use of the system as a vehicle for obtaining a clearer understanding of the role of intraneuronal mechanisms in brain function. ***