Super Greedy Trees

超级贪婪树

• 批准号: 10407442
• 负责人: Hemant Ishwaran
• 金额: $ 42.21万
• 依托单位: UNIVERSITY OF MIAMI SCHOOL OF MEDICINE
• 依托单位国家: 美国
• 财政年份: 2021
• 资助国家: 美国
• 起止时间: 2021-06-01 至 2026-05-31
• 项目状态: 未结题
• 来源: https://reporter.nih.gov/project-details/10407442
• 关键词: Big Data; Computer software; Data; Data Analyses; Machine Learning; National Institute of General Medical Sciences; Performance; Research Personnel; Structure; Survival Analysis; Time; Trees; base; classification trees; open source; regression trees; unsupervised learning

Project Summary/Abstract We identify critical weaknesses with Classification and Regression Trees (CART), a widely used base learner for machine learning of big omic-data analysis, and propose to replace these with a fundamentally different type of base learner we call super greedy trees (SGT's). SGT's cut the space in a fundamentally different manner, resulting in a richer partition structure with provable consistency and superior empirical performance. The project will develop a unified SGT framework for big data analysis using machine learning including the treatment of time varying covariate survival analysis, unsupervised learning, highly imbalanced data and multivariate regression. The SGT framework will be deployed within scalable and extensible open source software that will allow NIGMS researchers to deploy them to deal with their challenging big data problems.

项目概要/摘要 我们通过分类和回归树 (CART)（一种广泛使用的基学习器）识别关键弱点 用于大组学数据分析的机器学习，并建议用根本不同的类型替换它们 我们称之为超级贪婪树（SGT）的基础学习器以一种根本不同的方式切割空间， 从而产生更丰富的分区结构，具有可证明的一致性和优越的经验性能。 将开发一个统一的 SGT 框架，用于使用机器学习进行大数据分析，包括时间处理 变化协变量生存分析、无监督学习、高度不平衡数据和多元回归。 SGT 框架将部署在可扩展和可扩展的开源软件中，该软件将允许 NIGMS 研究人员部署它们来处理具有挑战性的大数据问题。