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 (e.g. via physics-informed models) 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.
Ongoing PhD supervision
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)
Completed PhD supervision
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)
Publications
2024
Rda-Inr: Riemannian Diffeomorphic Autoencoding Via Implicit Neural Representations. Dummer, S., Strisciuglio, N. & Brune, C.Gender Privacy Angular Constraints for Face Recognition, 1-1 (E-pub ahead of print/First online). Rezgui, Z., Strisciuglio, N. & Veldhuis, R.https://doi.org/10.1109/TBIOM.2024.3390586Enhancing Soft Biometric Face Template Privacy with Mutual Information-Based Image AttacksIn 2024 IEEE/CVF Winter Conference on Applications of Computer Vision Workshops (WACVW), Article 10495713 (pp. 1141-1149). IEEE. Rezgui, Z., Strisciuglio, N. & Veldhuis, R.https://doi.org/10.1109/WACVW60836.2024.00124Fourier-basis Functions to Bridge Augmentation Gap: Rethinking Frequency Augmentation in Image Classification. ArXiv.org. Vaish, P., Wang, S. & Strisciuglio, N.https://doi.org/10.48550/arXiv.2403.01944Regressing Transformers for Data-efficient Visual Place Recognition. ArXiv.org. Leyva-Vallina, M., Strisciuglio, N. & Petkov, N. 2023
Quantifying white matter hyperintensity and brain volumes in heterogeneous clinical and low-field portable MRI. ArXiv.org. Laso, P., Cerri, S., Sorby-Adams, A., Guo, J., Mateen, F., Goebl, P., Wu, J., Liu, P., Li, H., Young, S. I., Billot, B., Puonti, O., Sze, G., Payabavash, S., DeHavenon, A., Sheth, K. N., Rosen, M. S., Kirsch, J., Strisciuglio, N., … Iglesias, J. E.https://doi.org/10.48550/arXiv.2312.05119What do neural networks learn in image classification? A frequency shortcut perspectiveIn 2023 IEEE/CVF International Conference on Computer Vision (ICCV), Article 10378114 (pp. 1433-1442). IEEE. Wang, S., Veldhuis, R., Brune, C. & Strisciuglio, N.https://doi.org/10.1109/ICCV51070.2023.00138DFM-X: Augmentation by Leveraging Prior Knowledge of Shortcut LearningIn 2023 IEEE/CVF International Conference on Computer Vision Workshops (ICCVW), Article 10350684 (pp. 129-138). IEEE. Wang, S., Brune, C., Veldhuis, R. & Strisciuglio, N.https://doi.org/10.1109/ICCVW60793.2023.00020Riemannian Shape Manifold Learning with Applications to Biological Data. Dummer, S., Strisciuglio, N. & Brune, C.DFM-X: Augmentation by Leveraging Prior Knowledge of Shortcut Learning. ArXiv.org. Wang, S., Brune, C., Veldhuis, R. & Strisciuglio, N.https://doi.org/10.48550/arXiv.2308.06622Defocus Blur Synthesis and Deblurring via Interpolation and Extrapolation in Latent Space. ArXiv.org. Mazilu, I., Wang, S., Dummer, S., Veldhuis, R., Brune, C. & Strisciuglio, N.https://doi.org/10.48550/arXiv.2307.15461What do neural networks learn in image classification?: A frequency shortcut perspective. ArXiv.org. Wang, S., Veldhuis, R., Brune, C. & Strisciuglio, N.https://doi.org/10.48550/arXiv.2307.09829Data-efficient visual place recognition with graded similarity supervision. Leyva-Vallina, M., Strisciuglio, N. & Petkov, N.https://openaccess.thecvf.com/content/CVPR2023/papers/Leyva-Vallina_Data-Efficient_Large_Scale_Place_Recognition_With_Graded_Similarity_Supervision_CVPR_2023_paper.pdfRSA-INR: Riemannian Shape Autoencoding via 4D Implicit Neural Representations, 1-26. ArXiv.org. Dummer, S., Brune, C. & Strisciuglio, N.https://arxiv.org/abs/2305.12854Rda-inr: Riemannian Diffeomorphic Autoencoding via Implicit Neural Representations. ArXiv.org. Dummer, S., Strisciuglio, N. & Brune, C.https://doi.org/10.48550/arXiv.2305.12854 Research profiles
Scan the QR code or