About the project
Tensors, multi-dimensional arrays, are fundamental data structures in various scientific domains, including image analysis, neuroscience, the analysis of social network data, and modern genomics. However, in many real-world applications, tensors are only partially observed due to sensor limitations, missing data, or privacy concerns. Accurate estimation and recovery of these partially observed tensors is crucial for taking informed decisions and advancing our understanding of complex phenomena, and there is a pressing need for novel methods that can handle high-dimensional functional, sparse, and noisy data effectively.
In this project, we will develop knowledge-informed ML methods for analyzing tensor data including actionable structure such as large numbers of smooth curves, or spatial information. The idea is to combine efficient data-driven methods (Variational Empirical Bayes methods), in conjunction with latent factor models and tensor recovery methods, with functional data analysis techniques to develop a modeling adequate and efficient framework for analyzing large cubic data sets (fMRI, spatial transcriptomics, hyperspectral imaging, and other sequenced-based assays).
Project team
-
Researcher William Denault, Oslo University Hospital, Oslo Centre for Biostatistics & Epidemiology (OCBE)
-
Associate Professor Valeria Vitelli, University of Oslo, Institute of Basic Medical Sciences
-
Professor Tero Aittokallio, University of Oslo, Institute of Basic Medical Sciences
-
Professor Sylvia Richardson, University of Cambridge, MRC Biostatistics Unit
-
Associate Professor Aliaksandr Hubin, University of Oslo, Department of Mathematics