复杂网络体系下脑联想记忆功能网络的稀疏拓扑择优与建模研究

Associative memory modeling human intelligence as information storage and recovery is a hot research issue in cognitive neural computing, and receives widely application in the field of artificial intelligence and pattern recognition. But current models built on fully connected network focus on algorithm realization under fixed structure, therefore they are lack of biological modeling background and self-organized learning flexibility of the network topology and thus the intelligent level of these models is limited. Complex network focuses on the relationship between topology and function of the network, and it's a brand-new method of complex system investigation. Associative memory model is a dynamical nonlinear complex system in essence, small-word and scale-free characteristics are universal phenomenon in biological neural system. Therefore, from these points of view, it's a novel and feasible manner to study associative memory through complex network ideology. In this project, we utilize complex network ideology to research the influence of the neurons' sparse interconnection style on the network performance, i.e. topology versus function. Starting from the network topology, we do detailed theoretical analysis through non-equilibrium statistical method and also complementary numerical investigation. In the view of limited energy consumptions as in the human brain, we derive optimal synaptic dilution strategy under the constraint of limited energy consumptions on fully connected network, and research a heuristic annealed topological preferential mechanism for the network construction. Novel associative memory models built on complex networks are to be investigated so as to throw new light on the explanation of relationship between the intelligent complexity and the topology complexity. This project develops both the research contents and techniques for associative memory model, and it's of great significance for providing intelligent biomimetic computational model in the field of artificial intelligence and pattern recognition.

模拟生物脑神经认知系统的联想记忆机制，是仿生智能信息处理研究中的关键科学问题之一。建立在全互联结构上的联想记忆模型，只关注固定结构下学习算法的实现，缺乏对于生物脑神经系统建模的合理性，且不具备网络结构自组织学习的柔性，智能水平有限。本课题以探索新型联想记忆模型的实现方式为研究目标，借鉴生物脑神经系统中普适存在的复杂网络特性，以复杂网络结构与功能间关系的研究为切入点，使用非平衡态统计分析方法研究联想记忆在复杂网络拓扑结构上涌现的动力学行为，从理论分析、数值仿真及两者的一致性验证等方面研究复杂网络稀疏互联方式对其联想记忆性能的影响；并在此基础上，研究基于启发式退火拓扑择优机制的网络稀疏结构自组织学习方法，构建具有脑神经生物学依据的复杂网络小世界及无标度联想记忆模型。本课题的研究成果将为解释脑中智能复杂性与拓扑复杂性之间的联系提供理论与模型支持，在面向模式识别的仿生智能信息处理领域具有应用价值。

揭示并模拟大脑神经认知系统的学习机理是研究新一代仿生智能信息处理系统的关键。联想记忆网络模拟人脑信息存储及回忆机制，具有对含噪及不完全信息的鲁棒处理能力，因而在人工智能、模式识别等领域获得了广泛的研究与应用。建立在全互联结构上的联想记忆模型，只关注固定结构下学习算法的实现，缺乏对于生物脑神经系统建模的合理性，且不具备网络结构自组织学习的柔性，智能水平有限。本项目以探索新型联想记忆模型的实现方式为研究目标，借鉴复杂网络结构与功能间关系研究方法以及生物脑神经系统中普适存在的复杂网络性质，从网络体系结构角度出发，神经元稀疏互联方式入手，使用非平衡态统计分析方法从理论分析、数值仿真及两者的一致性验证等方面系统的研究了具有稀疏互联结构的联想记忆神经网络的性能及其动力学演化行为，并在此基础上研究了两类代谢能量约束准则下基于启发式退火拓扑择优机制的网络稀疏化策略，实现了联想记忆网络拓扑结构的自组织、动态择优学习。同时，模拟生物脑神经系统中广泛存在的复杂网络小世界效应和无标度特性，构建了基于启发式退火拓扑择优机制的复杂网络体系联想记忆模型，实现了联想记忆由“用数值隐含表示记忆模式”向“用结构连接表示记忆模式”的转变，为解释生物脑神经系统在生理发育期间变结构的网络动态调整现象以及脑中智能复杂性与拓扑复杂性之间的联系提供理论与模型支持。项目总体实施按照“生理原型背景支持分析->理论模型建立与性能分析->网络模型构建与算法实现研究”的主线逐层开展研究工作，拓展了现有联想记忆的研究内容和研究方法，充实和丰富其应用领域，在探索新型仿生智能计算模型和面向模式识别等研究领域具有理论意义与应用价值。