Distributional Data Representation Search for Multi-modal Machine Learning
The LungCAIRE project aims to improve the lung cancer relapse prediction through multi-modal patient profiling, including histopathological images, multiplex-immunofluorescence analyses, as well as gen and protein expressions, for finding indicative biomarkers. In this context, this subproject explores different data representations for multi-modal input data, data-centric training procedures, and distributional data representation search. We study semi-supervised learning including the summarization of data distributions, modality-specific correlations, and learned sampling and augmentation for facilitating the efficient and scalable training of high-capacity models.
Research Group Lead
Doctoral Researcher