报告题目：Consensus learning using distance preserving embedding methods
报告人：Dr. Andreas Nienkötter (Department of Computer Science，University of Münster)
Learning a consensus object is a core problem in machine learning and pattern recognition, where the combination of several approximate results into one prototype object can lead to a more accurate result with less outliers. For example, using the k-means clustering algorithm several times on the same dataset can lead to many similar but different results that can be combined into one more robust consensus result. A commonly used distance based approach for consensus learning is the so called generalized median. In many domains, the generalized median has a inherent high computational complexity (typically NP-hard). Approximate solutions for many types of objects can be found using the distance preserving embedding framework. In this talk I will present this framework including examples on different types of datasets. I will show that the framework performs well on a wide range of domains, leading to an accurate consensus computation in a short time. In addition, this framework is easily adaptable to other types of distance based optimization problems, as I will show with the related closest string problem.
Andreas Nienkötter received his Masters degree in Computer Science from the University of Münster, Germany in 2015. Since then, he is a PhD student in Prof. Xiaoyi Jiangs research group for Pattern Recognition and Image Analysis (PRIA) at the University of Münster. His research interests include consensus learning using the generalized median, vector space embedding methods and dimensionality reduction methods.