Structure Representation in Sub-spaces
In this research, we study structure detection in high-dimensional settings. Data in high dimensions reveals complicated and challenging properties. To tackle some of the challenges, we investigate how the underlying structures appear and evolve at different subspaces. For example, in image or gene expression data, some interesting patterns might be hidden in a subspace, whereas their detection from the original high-dimensional data is difficult, due to e.g. curse of dimensionality.
We first develop an efficient method to automatically identify the relevant subspaces. Then, for each selected subspace, we propose an appropriate data representation. Finally, we analyze several ways of combining the structures detected from different subspaces. Our approach will yield an automatic pattern discovery principle adaptive with respect to the context of the data at hand in the spirit of ensemble or deep representation methods. We will investigate the method on several standard and Xerox real-world datasets.
We are looking for a motivated PhD student/candidate in computer science or applied mathematics with a strong background in machine learning and statistics. Strong skills in implementation and programming with Python or Matlab are necessary.
XRCE provides a very international, motivating and innovative research environment. The intern will have the chance to perform top-level research on the theory and application of machine learning, as well as experience close collaboration with other researchers, engineers and the Xerox business team.
To submit an application, please send your CV and cover letter to both email@example.com and firstname.lastname@example.org. Please specify "Internship:Structure Representation in Sub-spaces " in your subject line.