2 PhD positions on Data-Efficient Foundation Models for Vision

The Data Management and Biometrics (DMB) group at the University of Twente is seeking two PhD candidates to join the research team of Dr. Nicola Strisciuglio for the NWO-VIDI project “ReVision: Robust and Data-Efficient Vision Foundation Models”.

About the project - ReVision
Foundation models in computer vision currently rely on massive datasets and brute-force scaling. This leads to high data requirements, hidden biases, limited accessibility, and dependency on corporate-controlled resources. ReVision aims to develop data-efficient and reliable training strategies for vision foundation models, reducing the need for large datasets and improving robustness.

The project will explore strategies related to 1) Object-attribute compositionality to replace exhaustive data requirements with structured concept learning, 2) Bias detection and machine unlearning to identify and mitigate bias and shortcuts early in training, and 3) Perceptual and conceptual priors to design self-supervised objectives that capture continuous similarity rather than binary contrastive notions.

By embedding structure and prior knowledge into the training process, ReVision aims to break the dependency on uncontrolled large-scale datasets and enable broader, more transparent development of vision foundation models.

About the PhD positions:
We offer two fully-funded 4-years PhD positions, with specific focus on Compositional Learning (PhD position 1) and Self-supervised Learning with priors (PhD position 2). The ReVision team will share efforts on integrating machine unlearning techniques to remove biases and shortcuts during model training.

While applying, you are asked to motivate the choice of (one or both) the position(s) you are applying for.