基于神经信息的稀疏深度认知建模及驾驶行为验证

Visual cognition is one of the most important functions in human brain, which is related to its language and culture. This project explores the model about visual cognition with Chinese culture based on neuroinformatics. We investigate multiple time series analysis techniques to analyze hidden cognitive states in the human brain. Spectral modularity approach is proposed to reconstitute the networks sparsely. In our sparse hierarchical deep model, there is a large-scale network structure at the top layers while there are the distributed multi-agent systems with time-varying at its low layers. At the top layer of the model, there is a large-scale network structure with top-down guidance learning, while the smaller scale is associated with a distributed multi-agent system through the bottom-up selective learning at its bottom layer. The agents play an important role in hidden chains to compile the modularized outputs. The system is developed with self-learning and emergence. Our model will be derived mathematically by formalization methods. And it would be verificated in driving behavior analysis systems. The study will be helpful to design some artificial cognitive systems to provide new computing models and approaches.

视觉认知是人脑的重要功能，与语言文化习惯密切相关。本项目在神经信息的基础上研究汉语文化背景下的视觉认知建模，结合脑结构和功能以及视觉认知实验，给出多元时序隐概映射方法分析脑区激活、负激活、抑制强度，时序化认知状态，提出谱分模块化和模块内外稀疏重建方法重构认知动态网络，对模块分化和节点配性系数进行测定，运用层次因果关系分析法和粗糙集多知识约简法解析其动态相关性。所提出的模型，在较大尺度的上层为模块化深度网络，每个模块下层是具有较小尺度的分布式多代理系统，其隐链模式的激活态随时间变换，表现出时变性；模型在群智稀疏深度学习过程中，上层在外显学习时使用下层获得的内隐输出，下层在内隐学习时使用上层获得的外显反馈，系统上层采用自上而下的指导性学习和下层采用自下而上的选择性学习模式，具有涌现性和自学习特征。最后对模型进行形式化推演，并在驾驶行为分析中验证。该项研究有望为设计智能系统提供新的思路。

视觉认知是人脑的重要功能，与语言文化习惯密切相关。本项目在神经信息的基础上研究汉语文化背景下的视觉认知建模，结合脑结构和功能以及视觉认知实验，给出多元时序隐概映射方法分析脑区激活、负激活、抑制强度，时序化认知状态，提出谱分模块化和模块内外稀疏重建方法重构认知动态网络，对模块分化和节点配性系数进行测定，运用层次因果关系分析法和粗糙集多知识约简法解析其动态相关性。所提出的模型，在较大尺度的上层为模块化深度网络，每个模块下层是具有较小尺度的分布式多代理系统，系统上层采用自上而下的指导性学习和下层采用自下而上的选择性学习模式，具有涌现性和自学习特征。最后对模型进行形式化推演，并在驾驶行为分析中验证。该项研究有望为设计智能系统提供新的思路。