From July 3rd to July 10th, I attended the VISUM summer school online. Originally planned to take place in Porto, it was moved online due to the pandemic.
The itinerary included lectures, AI shots, and a main competition. The lectures covered topics like machine learning, deep learning, information security, and AI. The AI shots were short 30-minute talks by industry professionals on AI use cases in their work.
Lecture Highlights
Jaime Cardoso (INESC TEC and the University of Porto FEUP) provided a detailed lecture on computer vision (CV) and machine learning (ML), covering deep learning, artificial intelligence, and their history.
One highlight was professor Johan Suykens’ lecture on Deep Learning and Kernel Machines. He discussed function estimation, LS-SVM, kernel spectral clustering, restricted kernel machines, and generative models.
Pascal Mettes (University of Amsterdam) gave an engaging talk on action recognition, discussing video representation, zero-shot learning, self-supervised learning, and graph neural networks for action understanding.
Professor Nicolas Courty’s talk on Optimal Transport in Deep Learning was equally intriguing, diving into Wasserstein Loss, domain adaptation, and Monge mapping.
Professor Marta Gomez-Barrero delivered an insightful talk on Information Security, covering cryptography, biometrics, user authentication, and a case study on Cancelable Biometrics Based on Bloom Filters [2].
Mauricio Reyes (University of Bern) discussed Interpretability in ML for Medical Imaging, differentiating between explainability and interpretability and emphasizing their importance in medical applications.
The Competition
The most exciting part was the competition. From 70+ participants, we formed teams of three to work on a fish detection problem. Using methodologies like YOLO, Faster R-CNN, and transfer learning with pre-trained models, the winning teams achieved results close to the state of the art.
References
- Suykens, J. A. K., et al. “Least squares support vector machine classifiers: a large scale algorithm.” Proceedings of the European Conference on Circuit Theory and Design, Vol. 10, 1999.
- Rathgeb, Christian, et al. “Towards cancelable multi-biometrics based on bloom filters: a case study on feature level fusion of face and iris.” 3rd International Workshop on Biometrics and Forensics (IWBF 2015). IEEE, 2015.