Research
Research
My research in Computer Vision and Machine Learning revolves around the problems of generalization and robustness to changing environments and perturbations of the input data. I focus on identifying and mitigating the sources of bias in training data, while inducing good bias via expert and prior knowledge in the training process, model architectures or labels.
I address the understanding of the training data characteristics and the learning dynamics of computer vision models, to identify sources of bias and counteract them. I investigate the use of prior knowledge, in the form of improved labeling procedures to better exploit data semantics, in the design of novel architectural elements or to steer the learning process for the training of data-efficient models, taking into account aspects related to training time and energy consumption. I envision efficient and robust computer vision models with embedded prior knowledge that are able to perform unbiased predictions, exploiting the semantics of the data rather than shortcut solutions, with reduced training time, size of the parameter space and energy consumption, with comparable performance to larger models trained with massive datasets and with higher computational requirements.
PhD students
Ongoing: \
- Shunxin Wang (University of Twente, Data Management and Biometrics group)
- Zohra Rezgui (University of Twente, Data Management and Biometrics group)
- Sven Dummer (University of Twente, Mathematics of Imaging and AI group)
- Melissa Tijink (University of Twente, Data Management and Biometrics group)
- Peter van der Wal (University of Groningen, Information systems group)
Graduated: \
- Dr. Maria Leyva Vallina (University of Groningen, The Netherlands - 2023)
- Dr. Virginia Riego del Castillo (University of Leon, Spain - 2022)
- Dr. Vincenzo Vigilante (University of Salerno, Italy - 2021)
Open student projects
You can find open projects on the website of the DMB group.
I also welcome your project proposal in computer vision related to: \
- model robustness and representation learning
- compositionality
- data augmentation
- shortcut learning and de-biasing
- image retrieval
with application to:
- robotics, scene understanding
- face image analysis
- microscopy images
- medical images