Konstantinos Kontras

Konstantinos Kontras

Multimodal Machine Learning

PhD Researcher KU Leuven

Welcome, I am Konstantinos and a member of Biomed STADIUS at ESAT, KU Leuven in Belgium as a PhD researcher. There along with my supervisors Prof. Maarten De Vos , Dr. Christos Chatzichristos and Prof. Johan Suykens , we are interested in Multimodal Fusion through learning processes.

Our goals are to find the best ways to allow modalities to combine information from each other, tolerate the absence of a modality subset and through multiple sources enhance our model calibration.

Before joining Biomed, I was a part of PSI/Visics group as a master student at the same building working on uncertainty for autonomous vehicles. Bayesian Model Average approximations and the CityScapes dataset was combined to create a pixel-level Active Learning schema for semantic segmentation. I had the chance to be supervised with Ir. Mark Proesmas , Ir. Jonas Heylen and Prof. Luc Van Gool

News

Nov 2023: Recently submitted to CVPR 2024 doubting the benefits of multimodal learning from scratch. More to come with the reviews!
March 2023: Talk in the Health Tech Summit on the role of ML/AI on future health systems.
March 2023: Shared CoRe-Sleep: A Multimodal Fusion Framework for Time Series Robust to Imperfect Modalities soon to be published in IEEE TSNRE.

Publications

(2023). CoRe-Sleep: A Multimodal Fusion Framework for Time Series Robust to Imperfect Modalities . IEEE TSNRE 2024 . Cite
(2023). Is there benefit in training multimodal models from scratch? CVPR 2024 . Cite

Contact