PhD position on Tensor Compression for building Machine Learning Models

Context: Artificial Intelligence has revolutionized the way images are processed. During the last years, spectral imagers (e.g., hyperspectral, CT) have become commercially available, which offer in particular great added for quality control in agrofood. However, spectral imaging devices are burdened by the complexity and high dimensionality of generated data compared to classical imagers. In addition, machine learning models typically require large, labelled datasets which are generally unavailable for spectral images. There is thus a need for efficient learning strategies. The objective of this project is to develop tensor-based compression algorithms for efficiently building machine learning models for high-dimensional, spectral images in agrofood applications.

In this research work, you will first evaluate the performance of tensor decomposition and compression when applied to spectral data and fed into a machine learning pipeline. Subsequently, you will go well beyond the state of the art and develop novel methods that adapt the tensor decomposition to align better with the machine learning goal (e.g., classification, regression, clustering).

This project is in collaboration with the DTAI research unit of the Computer Science department and the MeBioS division of the Biosystems department.