Project descriptionThe development of reliable Machine Learning algorithms for solving image-to-image problems like in medical image acquisition, requires a paradigm shift. On the algorithm development itself, there is a shift from theory-based models to data-based models or a combination of those. The new algorithms/models (often deep neural networks) outperform traditional methods, but they are notoriously inefficient to develop due to the trial-and-error nature of model development. Furthermore, they are also hard to trust and interpret due to the lack of explanation of the results. Visual Analytics solutions have shown potential for the explanation of ML models and their performance. The PhD in this vacancy will propose new visual analytics methods for ML models that are used in the context of medical imaging acquisition. In collaboration with ML researchers, the PhD will aim at providing the research community, industry, and clinical end users with visual analysis strategies to analyze, interpret, and improve their ML models and training data.
The project will be developed within the visualization cluster under the supervision of Prof.dr. Anna Vilanova and Dr. Nicola Pezzotti (Philips- part time TU/e) and in strong collaboration with experts in Machine Learning for acquisition purposes from the Electrical Engineering department at TU/e Dr.ir. Ruud van Sloun (TU/e - part time Philips).
The visualization cluster (
https://research.tue.nl/en/organisations/visualization) at TU/e has a strong track record in visualization and visual analytics for ML models and high-dimensional data. It has generated several award winning contributions at major visualization conferences (IEEE VIS, IEEE InfoVis, IEEE VAST, EuroVis); several successful start-up companies (MagnaView, Process Gold and SynerScope); and a number of techniques that are used on a large scale world-wide.
The visualization cluster participates actively in the newly created Eindhoven AI System Institute (
EAISI) and Data Science initiatives at TU/e.