Deep Learning for Density Estimation and Image Reconstruction

Victoria University of Wellington is internationally acclaimed for several key areas of artificial intelligence and nuclear magnetic resonance.

This research proposes a radical improvement in a broad range of imaging systems (including radar, radio telescopes, X-ray CT, MRI and X-ray crystallography) by combining a detailed understanding of the physics of the imaging systems with the most recent advances in machine learning techniques. This combination enables the effective use of prior information regarding the objects being imaged, so that less information is required from the imaging system. The resulting images will have dramatically improved accuracy and resolution, while requiring much less image acquisition time (leading to, e.g., significantly reduced cost and exposure to ionising radiation).

The physics of imaging processes are well understood, and can be accurately modelled. It is relatively easy to predict the measurements y that an imaging system will produce when imaging some given object x. However deducing x from y is a much more difficult inverse problem. Inverse problems are usually solved by the aid of generic constraints such as limits on total variation, sparseness or assumptions of partial separability of temporal and spatial components, non-local self-similarity, or low-rank. Domain specificity is an important aspect missing from even some very recent approaches. Domain-specific prior information is more difficult to utilise, but recent innovations in machine learning make it feasible to model this information.

We expect you to have a Master’s degree in electrical, or mechanical engineering, or physics with excellent grades. You will need a strong background in signal processing and machine learning, including competence in programming using the relevant libraries.
As part of a larger project team you will work, alongside expert full time researchers with strong backgrounds in machine learning and in imaging. Our previous students have sprung from a beginning with us to high-flying industry and academic careers.