CAREER: Advancing the Theory and Practice of Meta-Learning with Applications in Physics

职业：通过物理学应用推进元学习的理论和实践

• 批准号: 0448542
• 负责人: Ricardo Vilalta
• 金额: $ 50万
• 依托单位: University of Houston
• 依托单位国家: 美国
• 项目类别: Continuing Grant
• 财政年份: 2005
• 资助国家: 美国
• 起止时间: 2005-01-15 至 2010-12-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=0448542&HistoricalAwards=false
• 关键词: CAREER; Advancing; Theory; Practice; Meta

This project focuses on the field of meta-learning by investigating the relation between learning mechanisms and the tasks and domains where these mechanisms are applicable. An important activity of this project is to propose effective data structures and meta-features that characterize example distributions. We plan to extend model-based and information-theoretic characterizations, and to exploit multivariate density estimation techniques to generate a concise mapping of an example distribution. Results from this project will be used in the design and construction of meta-learning assistants that will be able to provide automatic and systematic user guidance for model selection and model ranking.This project is multidisciplinary in nature, with a particular emphasis on physics and astronomy. The principal area of application lies in the classification of Mars landscapes from geo-morphological features.A major goal of this project is to combine research and educational strategies aimed at the establishment of a Pattern Classification and Machine Learning Laboratory at the University of Houston. This laboratory will be used to analyze, learn, and perform predictions using real scientific data. It will also be used to help local youth, especially women and minorities, acquire a deep appreciation for science. Through the formation of science clubs and learning communities among high school and undergraduate students, this project strives to have a strong impact in fostering interest among young students in choosing scientific or technological paths as their professional career.

本项目的重点是元学习领域，通过研究学习机制与这些机制适用的任务和领域之间的关系。该项目的一个重要活动是提出有效的数据结构和元特征，以表征示例分布。我们计划扩展基于模型和信息理论的表征，并利用多元密度估计技术生成一个简洁的映射的例子分布。该项目的成果将用于设计和构建元学习助手，该助手将能够为模型选择和模型排名提供自动和系统的用户指导。该项目的一个主要目标是将联合收割机研究与教育战略结合起来，以便在休斯顿大学建立一个模式分类和机器学习实验室。这个实验室将用于分析，学习，并使用真实的科学数据进行预测。它还将用于帮助当地青年，特别是妇女和少数民族，获得对科学的深刻理解。通过在高中生和本科生中组建科学俱乐部和学习社区，该项目力求在培养青年学生选择科学或技术道路作为其职业生涯的兴趣方面产生重大影响。