格式塔规律的几何推理关键技术研究

项目名称： 格式塔规律的几何推理关键技术研究

项目编号： No.61273363

项目类型： 面上项目

立项/批准年度： 2013

项目学科： 自动化技术、计算机技术

项目作者： 文贵华

作者单位： 华南理工大学

项目金额： 80万元

中文摘要： 数据挖掘和机器学习等领域对维数灾难、数据噪音等问题还没有十分有效的解决算法，与此相反，人类却具有处理这类问题的良好能力。人类在高维小样本情况下仍然能够很好地分辨不同类别的样本，在同样的噪音环境下，也仍然能够识别出模糊的图像，因此我们需要从人类的认知规律中学到更多的东西。本立项研究刻画人类认知规律的格式塔（Gestalt）理论，将格式塔规律几何化，解决格式塔规律的几何推理问题，包括类比物理学规律建立格式塔规律的计算模型，利用微分拓扑、模糊拓扑等方法从数据集中自动构造符合格式塔规律的认知几何图形，构造认知几何图形知识库，研究认知几何图形的推理方法，研究认知几何图形的评价等关键问题，它们构成了有机整体，以实现多种格式塔规律的综合推理，解决复杂认知几何图形的识别、推理和评价问题。最后将其应用到机器学习和数据挖掘领域，提出新的原理和方法，解决这些领域的一些困难问题，验证格式塔几何推理系统的有效性。

中文关键词： 机器学习；认知规律；几何推理；数据挖掘；格式塔

英文摘要： In information domains such as data mining and machine learning,there are not very effective approaches to deal with the curse of dimensionality, the data noise, and the imbalance problem. However,Human being has the good ability to deal with these issues. Human being can distinguish samples well in the high dimensionsmall space and correctly identify a fuzzy image in the noisy enviroment.Thus we need to learn something more from the human being for designing new machine learning and data mining approaches. This project aims to do research on modelling the laws of cognitive Gestalt theory by geometric reasoning framework. The research items include modelling Gestalt laws based on analogy of laws of physics to establish the computational models, applying the differential topology and fuzzy topology methods to automatically construct the geometry graphs from the high dimensional data in line with the Gestalt laws, establishing the knowledge base of cognitive geometry graphs, developing the cognitive geometry reasoning approaches, evaluating the constructed cognitive geometry graphs,and the other related key techniques. These research items constitute an organic whole to perform integrated reasoning of Gestalt laws so as to solve the recognition, reasoning, and evaluation on complex cognitive geometry graphs. Final

英文关键词： machine learning；cognitive laws；geometrical reasoning；data mining；Gestalt