CAREER: An Information-Theoretic Approach to Computational Learning with Applications

职业：计算学习的信息理论方法及其应用

• 批准号: 0093131
• 负责人: Javed Aslam
• 金额: $ 25万
• 依托单位: Dartmouth College
• 依托单位国家: 美国
• 项目类别: Continuing Grant
• 财政年份: 2001
• 资助国家: 美国
• 起止时间: 2001-07-01 至 2004-04-30
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=0093131&HistoricalAwards=false
• 关键词: CAREER; Information; Theoretic; Approach; Computational

Information theory is the science of quantifying, encoding, andextracting information. The research proposed in this careerdevelopment plan consists of work designed to strengthen thetheoretical foundations and empirical applicability of computationallearning through the development of a rigorous, information-theoreticframework for investigating this discipline. Specifically, aninformation-theoretic framework, based on communication theory, isproposed for investigating the probably approximately correct (PAC)model of machine learning. This framework is shown to be robust andextensible; new learning algorithms, general measures of performance,and analysis techniques are all proposed and demonstrated.In addition to providing a means for strengthening the theoreticalfoundations of computational learning, this framework also provides amechanism for developing and rigorously analyzing new learningalgorithms. New algorithms for hypothesis boosting, documentclassification, document filtering, and meta-search are all proposed.The proposed applications work has both theoretical and experimentalcomponents: all new algorithms are to be analyzed, implemented andtested using benchmark data.Teaching and education are an integral part of this career developmentplan. Both graduate and undergraduate students are introduced totopical research, both theoretical and applied, through the projectsproposed. New courses on learning and information retrieval,accessible to both graduate and undergraduate students, are alsoproposed.

信息论是对信息进行量化、编码和提取的科学。这项职业发展计划中提出的研究包括旨在通过开发一个严格的信息论框架来研究这一学科，以加强计算学习的理论基础和经验适用性的工作。具体地说，提出了一个基于通信理论的信息论框架，用于研究机器学习的可能近似正确(PAC)模型。该框架具有较强的健壮性和可扩展性，提出并展示了新的学习算法、通用性能度量和分析技术，不仅为加强计算学习的理论基础提供了手段，也为开发和严格分析新的学习算法提供了一种机制。提出了假设提升、文档分类、文档过滤和元搜索的新算法。提出的应用工作包括理论和实验两个部分：所有新算法都将使用基准数据进行分析、实现和测试。教学和教育是职业发展计划中不可或缺的一部分。通过提出的项目，研究生和本科生都被介绍到专题研究中，包括理论研究和应用研究。此外，还建议开设研究生和本科生均可使用的学习和信息检索新课程。