Effective Learning by Leveraging Supervised and Unsupervised Techniques

提案0534286“利用监督和无监督技术进行有效学习“PI：丹尼尔L。BoleyUniversity of Minnesota-Twin Cities摘要众所周知，在许多科学学科和网络上产生的数据爆炸需要开发有效的可扩展技术，以（半）自动化的方式挖掘这些数据中的知识。在方法的可扩展性和提取的知识质量之间存在着权衡。该项目旨在利用最强大的机器学习算法（例如支持向量机内核方法）与可扩展的聚类算法，以允许探索巨大的数据集，获得对这些方法的预期性能的估计，并扩展这些方法，以便它们可以学习隐藏在数据中的新的未预料到的概念。这项研究将产生更广泛的影响，使有效的，既定的机器学习技术在生物学和其他科学学科产生的大量数据集合中的使用成为可能，从而可能导致发现隐藏在数据中的新知识。

Proposal 0534286"Effective Learning by Leveraging Supervised and Unsupervised Techniques"PI: Daniel L. BoleyUniversity of Minnesota-Twin CitiesABSTRACTThe well-recognized explosion of data being generated in many scientific disciplines and on the Web necessitates the development of effective scalable techniques to mine this data for knowledge in a (semi)-automated way. There has been a trade-off between the scalability of methods and the quality of knowledge extracted. This project aims to leverage the most robust machine learning algorithms (e.g. support vector machine kernel methods) with scalable clustering algorithms to permit the exploration of enormous data sets, to derive estimates of the expected performance of such methods, and to extend such methods so that they can learn new unanticipated concepts buried within the data. This research will have broader impact by making possible the use of effective, established, machine learning techniques on the large aggregations of data being generated in biology and other scientific disciplines, thus, potentially leading to the discovery of new knowledge hidden within the data.