RUI: Novel Enhancements to the K-Means Clustering Algorithm

RUI：K-Means 聚类算法的新颖增强

• 批准号: 1117457
• 负责人: M. Emre Celebi
• 金额: $ 15.59万
• 依托单位: Louisiana State University Shreveport
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2011
• 资助国家: 美国
• 起止时间: 2011-07-01 至 2014-06-30
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=1117457&HistoricalAwards=false
• 关键词: RUI; Novel; Enhancements; Means; Clustering

Clustering is a crucial component of exploratory data analysis. This project aims to develop novel algorithms that address various shortcomings of k-means, the most widely used clustering algorithm. Specific objectives include (a) the development of initialization methods to address the sensitivity of k-means to the initial cluster centers; (b) the investigation of alternative distance measures to address the sensitivity of k-means to outliers; and (c) the development of practical acceleration methods. Innovations developed during this project should be readily applicable to a wide range of clustering algorithms. For example, initialization is crucial for most clustering algorithms. Furthermore, an effective initialization method can be used independently of k-means as a standalone clustering algorithm. K-means is often used as a subroutine in other learning algorithms. Therefore, development of acceleration methods for k-means is of great practical interest.The project is expected to make broader impacts on several fronts. At the international level, we intend to make significant contributions to the data mining literature by publishing in top-ranked journals. At the national level, we aim to enhance the competitiveness of the US by seeding the next generation of scientists. At the regional level, we hope to improve the quality of education in an EPSCoR state and contribute to the development of a diverse and skilled workforce. At the institutional level, we intend to improve the research environment and the curriculum of our Computer Science program. Finally, at the individual level, we hope to increase the participation of students from underrepresented groups in research and equip them with valuable skills including self-confidence, independent thinking/problem solving, and effective communication.

聚类是探索性数据分析的重要组成部分。该项目旨在开发新颖的算法来解决 k 均值（最广泛使用的聚类算法）的各种缺点。具体目标包括 (a) 开发初始化方法以解决 k 均值对初始聚类中心的敏感性； (b) 研究替代距离测量，以解决 k 均值对异常值的敏感性； (c) 开发实用的加速方法。该项目期间开发的创新应该很容易适用于各种聚类算法。例如，初始化对于大多数聚类算法至关重要。此外，有效的初始化方法可以独立于 k 均值作为独立的聚类算法使用。 K-means 通常用作其他学习算法中的子程序。因此，开发 k 均值加速方法具有重大的实际意义。该项目预计将在多个方面产生更广泛的影响。在国际层面，我们打算通过在顶级期刊上发表文章来为数据挖掘文献做出重大贡献。在国家层面，我们的目标是通过培养下一代科学家来增强美国的竞争力。在地区层面，我们希望提高 EPSCoR 州的教育质量，并为多元化和熟练劳动力的发展做出贡献。在机构层面，我们打算改善计算机科学项目的研究环境和课程。最后，在个人层面，我们希望增加来自弱势群体的学生对研究的参与，并赋予他们宝贵的技能，包括自信、独立思考/解决问题和有效沟通。