Tolerance-based Granular Computing Methods in Learning: Foundations and Applications

学习中基于容差的粒度计算方法：基础和应用

• 批准号: RGPIN-2019-04104
• 负责人: Ramanna, Sheela
• 金额: $ 1.68万
• 依托单位: University of Winnipeg
• 依托单位国家: 加拿大
• 项目类别: Discovery Grants Program - Individual
• 财政年份: 2022
• 资助国家: 加拿大
• 起止时间: 2022-01-01 至 2023-12-31
• 项目状态: 已结题
• 来源: https://www.nserc-crsng.gc.ca/ase-oro/Details-Detailles_eng.asp?id=743627
• 关键词: Tolerance; based; Granular; Computing; Methods

The primary significance of the proposed research accrues from a tolerance-based framework for different forms of knowledge representation and learning. This research bridges the gap between approximate and crisp forms of knowledge structures. Tolerance methods provide a formal foundation for almost solutions that are valid within some approximation, which is required for real world problems and applications. The utilization of various forms of set-based and graph-based approaches and nearness distance functionals of feature vectors make it possible to support approximate learning in three distinct areas: web information labelling, social networks and perception-based image and audio information. The proposed approach also explores computational challenges with big data in the three areas. A direct result of this research is its utility in a number of application areas. The novelty of the work is in: i) unifying theme of tolerance methods and nearness measures in distinct yet related areas, ii) establishing fundamental results of descriptive intersection based on regions and collections of regions and their interplay with classical spatial proximity, iii) unique approach to visual descriptive proximity leading to a broad spectrum of applications such as image analysis, iv) clustering based on Voronoi tessellations useful in efficient neighbourhood methods, v) change detection useful in visual surveillance, and vi) potential for technology transfer specifically with automated map production. This work will also help reveal that the representational structures viewed through the prism of tolerance have a great deal of fluidity and combined with a geometric view have the potential of integrating conceptual structures at different levels of granularity.

提出的研究的主要意义在于为不同形式的知识表示和学习提供了一个基于容忍度的框架。这项研究弥合了近似和清晰的知识结构形式之间的差距。容差方法为在某种近似范围内有效的几乎解提供了一个正式的基础，这是现实世界的问题和应用程序所需要的。利用各种形式的基于集合和基于图的方法以及特征向量的近距离函数，可以在三个不同的领域支持近似学习：网络信息标签、社会网络和基于感知的图像和音频信息。提出的方法还探讨了这三个领域的大数据计算挑战。这项研究的一个直接结果是它在许多应用领域的实用性。这项工作的新奇之处在于：i)在不同但相关的领域统一容忍方法和接近度量的主题，ii)建立基于区域和区域集合的描述性相交的基本结果及其与经典空间接近的相互作用，iii)独特的视觉描述性接近方法导致广泛的应用，如图像分析，iv)基于Voronoi细分的聚类，在有效的邻域方法中有用。V)在视觉监控中有用的变化检测，以及vi)技术转让的潜力，特别是自动化地图制作。这项工作还将有助于揭示，通过宽容棱镜观察的代表性结构具有很大的流动性，并与几何视图相结合，具有整合不同粒度级别的概念结构的潜力。