BITS: Reconfigurable and Multifunctional Behavioral Pattern Generators

我们建议开发新的理论和模型，通过理解和实现生物系统的动态可重构和多功能信息处理体系结构来扩展现有的计算框架。我们将通过一个合作的跨学科研究计划来应对这一挑战，该计划专注于多功能神经机械组件及其在动物身上重新配置为多种行为模式。我们提出的研究的最终目标是抽象出最终可以在各种其他环境中应用的一般设计原则。具体地说，我们提出了以下四个紧密交织的实验和建模/理论项目：1)我们将对软体动物加州海兔的摄食系统进行详细的实验分析，该系统根据环境变化动态地重新配置其摄食行为，并通过其神经力学的多功能性来实现这一点。首先，我们将描述动物在远距离行为模式之间切换的条件。其次，我们将研究这些切换的神经和机械基础。2)我们将创建和分析行为模式切换模型，作为新设计原则的基础。首先，我们将构建互联的半马尔可夫模型来捕捉在生物系统中观察到的行为模式重新配置。其次，我们将致力于为工程系统开发一种系统的设计方法，并将其潜在地应用于机器人装配。3)我们将创建和分析多功能图案生成的模型，以确定一般设计原则。首先，我们将使用遗传算法来进化多功能神经模式生成器，这种生成器可以在不同的行为模式之间切换。其次，我们将对这些模型模式产生器的设计空间进行详细的研究。4)探索多功能神经模式产生器在模拟VLSI中的实现。首先，我们将在上面提出的实验和理论工作的基础上，开发紧凑、低功耗的模式发生器。其次，我们将研究噪声和分量失配对这些网络性能的影响。

EIA-0130773 Randall D. Beer-Case Western Reserve-Reconfigurable and Multifunctional Behavior Pattern Generators We propose to develop new theories and models that extend current computational frameworks by understanding and implementing the dynamically reconfigurable and multifunctional information processing architectures of biological systems. We will address this challenge through a collaborative interdisciplinary research program focusing on multifunctional neuromechanical components and their reconfiguration into multiple behavioral patterns in animals. The ultimate goal of our proposed research is to abstract general design principles that can eventually be applied in a variety of other contexts. Specifically, we propose the following four closely intertwined experimental and modeling/theoretical projects:1)We will undertake a detailed experimental analysis of the feeding system of the mollusk Aplysia California, which dynamically reconfigures its feeding behavior in response to changing environmental circumstances, and does so through the multifuntionality of its neuromechanics. First, we will characterize the conditions under which the animal switches between distant behavioral patterns. Second, we will examine the neural and mechanical basis of these switches.2)We will create and analyze models of behavioral pattern switching as the basis for new design principles. First, we will construct interconnected semi-Markov models to capture behavioral pattern reconfiguration observed in biological systems. Second, we will pursue the development of a systematic design methodology for engineered systems, with potential applications to robotic assembly.3)We will create and analyze models of multifunctional pattern generations in order to identify general design principles. First, we will use genetic algorithms to evolve multifunctional neural pattern generators that can switch between distinct behavioral patterns. Second, we will undertake a detailed study of the "design space" of these model pattern generators.4)We will explore the implementation of multifunctional neural pattern generators in analog VLSI. First, we will develop compact, low-power pattern generators based on the experimental and theoretical work proposed above. Second, we will study the effects of noise and component mismatch on the performance of these networks.