MECHANISMS OF ASSOCIATIVE MEMORY

联想记忆的机制

• 批准号: 6530917
• 负责人: PETER E LATHAM
• 金额: $ 11.45万
• 依托单位: UNIVERSITY OF CALIFORNIA LOS ANGELES
• 依托单位国家: 美国
• 财政年份: 2001
• 资助国家: 美国
• 起止时间: 2001-03-01 至 2004-02-29
• 项目状态: 已结题
• 来源: https://reporter.nih.gov/project-details/6530917
• 关键词: association learning; computer simulation; mathematical model; memory; model design /development; neural information processing

DESCRIPTION (Adapted from Applicant's abstract): Associative memory is a fundamental property of the nervous system. It allows us to retrieve memories from partial or corrupted input, and thus plays a critical role in processing and categorizing information. The mechanisms that underlie associative memory are not yet understood. The leading model, a theoretical model, proposes that the nervous system performs associative memory by implementing attractor networks. In such networks, input, in the form of patterns of action potentials, provides partial information about a memory; the dynamics of the network then drives the neural activity to an attractor - a stable state in activity space - that corresponds to a complete representation of the memory. While the attractor model is a valuable construct, it is an idealized one - there is a large gap between the model and real neuronal networks. Our goal is to determine whether the attractor model is viable for real neuronal networks. The proposal is divided into two parts. The first is to determine whether networks with biologically -realistic properties can act as attractor networks. We will use analytical approaches, primarily mean field theory, to determine, in the abstract, the conditions necessary for the existence of attractors, and we will assess whether biological networks can satisfy these conditions. Large-scale simulations of realistic networks will then be performed to verify our analytical findings. The second is to determine how attractors can be learned. We will apply patterned input to a network, and examine network behavior as a function of the synaptic learning rule and the patterns of input to the network. In sum, our goal is to determine whether attractor networks can be realized in biological neuronal networks, and, if so, to determine the input patterns that will result in their formation. Successful completion of this project will provide us with experimentally testable predictions, predictions that can serve as a guide for investigating attractor networks in the nervous system.

描述（改编自申请人的摘要）：关联记忆是一个 神经系统的基本属性。它使我们能够检索回忆 从部分或损坏的输入中，因此在处理中起着至关重要的作用 并分类信息。基于关联记忆的机制 尚未理解。领先的模型，一个理论模型，提出了 神经系统通过实现吸引子来执行联想记忆 网络。在这样的网络中，输入，以动作模式的形式 电势，提供有关内存的部分信息；动力学的动态 然后，网络将神经活动驱动到吸引子 - 稳定状态 活动空间 - 对应于内存的完整表示。 虽然吸引子模型是一种有价值的结构，但它是理想化的模型 - 模型和实际神经元网络之间存在很大差距。我们的目标是 确定吸引子模型是否对真实神经元网络可行。 该提案分为两个部分。首先是确定是否 具有生物现实主义特性的网络可以充当吸引者网络。 我们将使用分析方法（主要是均值现场理论）来确定， 抽象地，吸引子的存在所需的条件，以及 我们将评估生物网络是否可以满足这些条件。 然后将对现实网络进行大规模模拟以验证 我们的分析结果。第二个是确定吸引子如何 学会了。我们将向网络应用模式的输入，并检查网络 行为是突触学习规则和输入模式的函数 到网络。 总而言之，我们的目标是确定是否可以在 生物神经元网络，如果是的，则确定输入模式 将导致他们的形成。成功完成该项目将 为我们提供可以实验测试的预测，可以服务的预测 作为调查神经系统吸引力网络的指南。