Wilson Silva is an Assistant Professor in the Departments of Information and Computing Sciences and Biology at Utrecht University, as well as a Guest AI Researcher in the Department of Radiology at the Netherlands Cancer Institute. His research focuses on trustworthy artificial intelligence for life sciences, with particular emphasis on explainable AI, privacy-preserving machine learning, and enhancing generalizability in cancer research.
Before moving to the Netherlands, Wilson completed his PhD at the University of Porto/INESC TEC, where he conducted research within the Visual Computing and Machine Intelligence group under the guidance of Prof. Jaime Cardoso. During his PhD, he also spent time as a visiting researcher in Prof. Mauricio Reyes’ lab at the University of Bern in Switzerland. Wilson holds a master’s degree in Electrical and Computer Engineering from the University of Porto.

TBA

Yongshuo Zong is a final-year PhD student at the University of Edinburgh where he works with Prof. Timothy Hospedales. His research focus on multimodal learning, especially large vision-language models, with an emphasis on areas such as safety, robustness, in-context learning, and multi-image understanding. He also works in the area of AI for healthcare where he develops MEDFAIR, one of the most popular benchmarks for medical imaging fairness.

Large language models (LLMs) have become integral to daily life, powering numerous applications. Building on the foundation of LLMs, large vision-language models (VLMs) have been developed to seamlessly integrate visual and textual data, enabling advanced multimodal interactions with humans. In this talk, I will first provide an overview of the background of large language models and multimodal learning. I will then introduce key research challenges in advancing LLMs and VLMs, focusing on topics such as in-context learning, multi-image understanding, and long-context processing. Finally, I will discuss the challenges of deploying large vision-language models safely, discussing potential alignment strategies to ensure reliability and ethical use. This talk aims to give the audience an overview of the foundations, advancements, and current challenges in vision-language models, highlighting the latest research driving this field forward.

Dr. Fadi Boutros is a scientific researcher at the Fraunhofer IGD and a principal investigator at the National Research Center for Applied Cybersecurity ATHENE, Germany. Fadi received his Ph.D. in computer science from TU Darmstadt (2022) and a master's degree in "Distributed Software Systems" from TU Darmstadt (2019). Also, he is participating in the Software Campus program, a management program of the German Federal Ministry of Education and Research (BMBF). He authored and co-authored several conference and journal papers. His main research interests lie in the fields of biometrics, machine learning, and efficient deep learning. For his scientific work, he received several awards, including the CAST-Förderpreis 2019 award, the IJCB 2022 Qualcomm Audience Choice Award, and the 2022 EAB Biometrics Industry Award from the European Association for Biometrics (EAB) for his Ph.D. dissertation.

In this lecture on knowledge distillation (KD), we delve into its critical role in enhancing the performance of compact machine learning models, using face recognition as a prime example application. KD is a model compression technique that involves transferring knowledge from a large, complex teacher model to a smaller, more efficient student model. In the context of face recognition, KD has emerged as an effective approach to enhance the performance of lightweight models, ensuring they perform well in resource-constrained environments. This lecture will explore the principles of knowledge distillation and its application to face recognition tasks. We will discuss various distillation strategies, including traditional soft-label techniques and more advanced methods such as feature-based distillation. The lecture will also address the challenges specific to face recognition, such as handling large-scale datasets, managing intra-class variability and performing KD within a privacy friendly framework. The lecture further addresses the data requirements for effective distillation, highlighting the importance of high-quality labeled datasets and discussing techniques to simulate diverse and representative data distributions. Real-world examples from face recognition systems, like those used in security and authentication, will demonstrate how knowledge distillation bridges the gap between model efficiency and accuracy, enabling scalable deployment. Finally, the lecture will provide insights into the future of KD in face recognition, including integration with deep learning architectures and the potential for improving real-time performance on edge devices.

Giulia D'Angelo is currently an MSCA Postdoctoral Fellow at the Czech Technical University in Prague, focusing on neuromorphic algorithms for active vision. She obtained a Bachelor's degree in Biomedical Engineering at The University of Genoa and a Master's degree in Neuroengineering, during which she developed a neuromorphic system for the egocentric representation of peripersonal visual space at King's College London. She earned her PhD in neuromorphic algorithms at the University of Manchester, in collaboration with the Event-Driven Perception for Robotics Laboratory at the Italian Institute of Technology, where she proposed a biologically plausible model for event-driven, saliency-based visual attention. She was recently awarded the Marie Skłodowska-Curie Fellowship, through which she explores sensorimotor contingency theories for neuromorphic active vision algorithms.

Vision is an exploratory behaviour that relies heavily on the dynamic relationship between actions and sensory feedback. For any agent—whether animal or robotic—processing visual sensory input efficiently is crucial for understanding and interacting with its environment. The key challenge lies in selectively filtering relevant information from the constant stream of complex sensory data. This process, known as selective attention, is also driven by the intricate interplay between bottom-up and top-down mechanisms, which together organize and interpret visual scenes. I will explore how biologically plausible models for visual attention can enhance robotic interaction with the environment trying to understand the role of neuromorphic hardware in facilitating active vision and its limitations.

Cosmin Bercea is a postdoctoral researcher at the Technical University of Munich (TUM), working at the Chair for Computational Imaging and AI in Medicine (TUM School of Computation, Information and Technology) and the Chair for AI for Image-Guided Diagnosis and Therapy (TUM School of Medicine and Health). His research focuses on vision and multimodal learning in medicine, aiming to improve the early and accurate diagnosis of rare and complex medical conditions. He completed his PhD at TUM under the supervision of Prof. Julia Schnabel and Prof. Daniel Rueckert, where he developed robust machine learning algorithms for detecting a wide range of previously unseen anomalies in medical images. His work combines generative modeling, distribution learning, and AI-driven medical analysis to develop reliable and interpretable decision support systems in healthcare.

Diffusion models have emerged as a powerful generative framework in natural image processing, setting new benchmarks in realism and controllability. Their success has spurred increasing interest in medical imaging, where they offer transformative potential across the imaging pipeline—from acquisition and reconstruction to downstream analysis and diagnosis.
In this talk, I will introduce the fundamentals of diffusion models and the challenges they face when applied to medical imaging, particularly in solving inverse problems such as image reconstruction and denoising. I will then explore their role in synthetic data generation, where they not only enable augmentation for data-scarce settings but also facilitate counterfactual image generation—producing plausible variations of medical scans that help disentangle pathological from normal variations. A key aspect of this is conditioning diffusion models with language, enabling them to integrate domain knowledge and enhance interpretability.
Building on these generative capabilities, I will focus on how diffusion models can support anomaly detection and aid in diagnosing rare and complex diseases by learning structured representations of normative anatomy. These models enable more refined anomaly localization and characterization, improving diagnostic support in challenging cases. I will also discuss how advancements in guidance techniques help refine model outputs, ensuring clinical relevance and enhancing their practical usability in medical imaging. This talk will provide a comprehensive perspective on the evolving role of diffusion models in medical imaging, highlighting their potential and the challenges that lie ahead.

Professor at Intituto Quimico de Sarrià in Barcelona and researcher in the field of Semantic Technologies and Artificial Intelligence at the National Center for Supercomputing (BSC-CNS)- Spain, where she has led national and international projects applying this technology to Smart Cities, social exclusion and poverty, and cultural heritage. Coordinator of the European project Saint George on a Bike, recently completed, which tried to automate with AI techniques the understanding of the visual content of paintings from the 12th-18th centuries and whose results will be applied within the framework of a collaboration with the Prado Museum. As a parallel line of research, she was involved in a long-term collaboration with the Carlos III University of Madrid and the National Center for Epidemiology (CNE), whose purpose was to simulate and implement strategies to evaluate the evolution of infectious diseases such as COVID; The tool developed is one of the few used by the ECDC (European Center for Disease Prevention and Control) Modeling Hub. She is the author of more than 50 articles in prestigious conferences and international journals, has participated in invited conferences, has several patents, and has participated in standardization work. She has a PhD from the University of California, has been a PostDoc at MIT, and has worked for several years at IBM T.J.Watson.

In this talk we will explore the fundamentals of text-vision models, their different types of architecture, the different options and challenges, and some of their applications. We will also discuss fine-tuning of these models to avoid problems still existing when applying these methods to domain with “small” data – e.g. hallucinations. We will also present some examples of using semantics to guide the learning outcomes, which seems to make a difference in the performance of both unimodal and multimodal AI for small data domains.

Ph.D. Luís Rosado is a Senior Scientist at Fraunhofer Portugal AICOS. He received a Ph.D. in Biomedical Engineering (2018) from the Faculty of Engineering of the University of Porto, and a M.Sc. in Biomedical Engineering (2009) from Instituto Superior Técnico of the Technical University of Lisbon. His research focuses on applying computer vision and machine learning to various industry sectors, being currently engaged in developing solutions for health, viticulture, and the automotive industry. Current areas of interest include Generative AI, Data-centric ML, Explainable AI, and Industrial Visual Inspection.

Ivo Façoco earned his M.Sc. in Computer Engineering and Informatics from FEUP Faculty in 2023. He is currently a Scientist at the Intelligent Stems research group at the Fraunhofer Center for Assistive Information and Communication Solutions (AICOS). His research focuses on image synthesis, data quality assurance, and data-driven techniques to enhance machine and deep learning algorithms. He has contributed to projects in both the medical and industrial domains, with a particular emphasis on image-based applications.

This session will allow a case study demonstration of controllable strategies for generating realistic and representative synthetic retinal fundus images with diffusion models. With a hands-on component focused on enhancing data auditing of real and synthetic retinal fundus image datasets using the recently open-sourced pyMDMA library. This will motivate a discussion about the practical implications of synthetic health data, including its benefits, risks, and ethical considerations.

Hélder P. Oliveira received the Ph.D. degree in electrical and computer engineering from the University of Porto, Portugal, in 2013. He is working at the Computer Science Department of the Faculty of Sciences of the University of Porto as an Assistant Professor. He is also working as Senior Researcher at INESC TEC and at Visual Computing and Machine Intelligence Area. He co-authored 50+ international peer-reviewed papers, have one patent conceded (Europe, China, Japan) and 90+ works in international conferences. In total, these publications have attracted +1800 citations, with an h-index of 25 according to Google Scholar. He is the Director of the Master in Medical Informatics from University of Porto. Hélder Oliveira is member of IEEE and also of the Portuguese Association of Pattern Recognition (APRP), which was elected as member of the presidency in 2023. His research interests include Computer Vision and Machine Learning on biomedical fields.

In modern healthcare, especially in Cancer, medical imaging plays an important role throughout the clinical process, from diagnosis and treatment planning to surgery and follow-up research. Patient triage in acute and chronic care, imaging-guided interventions, and treatment planning optimization are now part of standard clinical practice in several subspecialties. Medical images normally face one or more of the following challenges during acquisition: low resolution (in the spatial and spectral domains); high-level of noise; low contrast appearance; presence of geometric deformations; and the presence of imaging artefacts. Finer spatial sampling, which can be achieved with longer acquisition times, can avoid the drawbacks of low resolution and low contrast. However, this also increases the probability of geometric transformations, such as patient movement, resulting in blurred images. Imaging artefacts also lead to challenging problems in medical image analysis. Learning methodologies could be also a way to help physicians in their daily tasks, for example when dealing with multiple data sources, data sparsity or providing information about visual findings using, for example, explainable methods. Our main activities are focused on interdisciplinary research, computational models on health data from multiple sources, models integrating clinical domain knowledge and transparent, causal and accurate algorithms for improving clinical outcomes.

Ph.D. Luís Rosado is a Senior Scientist at Fraunhofer Portugal AICOS. He received a Ph.D. in Biomedical Engineering (2018) from the Faculty of Engineering of the University of Porto, and a M.Sc. in Biomedical Engineering (2009) from Instituto Superior Técnico of the Technical University of Lisbon. His research focuses on applying computer vision and machine learning to various industry sectors, being currently engaged in developing solutions for health, viticulture, and the automotive industry. Current areas of interest include Generative AI, Data-centric ML, Explainable AI, and Industrial Visual Inspection.

The EU-funded AISym4MED project is creating a platform to support the development of digital solutions and services for medical applications by enabling innovators to develop, evaluate and improve AI systems in a secure and trustworthy environment. AISym4MED will facilitate the development of robust AI solutions in healthcare by leveraging synthetic data generation and auditing, ensuring secure data sharing to improve medical research and patient outcomes.

Luís Pessoa is currently an Assistant Professor at the University of Porto, group leader at the Centre of Telecommunications and Multimedia from INESC TEC, and member of the Steering Board of the 6G Smart Networks and Services (6G-SNS) initiative. His main research interests include reconfigurable intelligent surfaces and computer vision towards integrated communications, localisation and sensing in 6G.

In this talk I will introduce CONVERGE, a project that aims to create an innovative research toolset in line with the motto “view-to-communicate and communicate-to-view”. The toolset will comprise vision-aided large intelligent surfaces, vision-aided fixed and mobile base stations, a vision-radio simulator and 3D environment modeller, and machine learning algorithms for multimodal data integration, such as wireless communications signals, video/lidar streams, radio/radar sensing, and network traffic traces. The integration of all these domains will play a crucial role in shaping the next generation of wireless communication technologies and associated applications.

Kelwin Fernandes is passionate about discovering new ways of using artificial intelligence (AI) in businesses and has expertise in AI-based digital transformation and data-driven decision-making in organizations. He is the CEO of NILG.AI, an AI consultancy company that aims to become the standard for AI adoption, making any company highly efficient, ever-improving, and adaptable.

TBA

Susana Brás is a researcher at IEETA, University of Aveiro, with a background in Applied Mathematics and a PhD in Biomedical Sciences. Her research focuses on affective computing and the responsible application of AI, exploring human behavior and environmental contexts to enhance quality of life. She addresses challenges such as class imbalance and missing data. With experience in student supervision, scientific publishing, and project management, Susana is a strong advocate for clear scientific communication. Passionate about promoting gender equity in tech, she is actively involved in initiatives like Geek Girls Portugal.

In an era where data drives decisions, we often focus on volume, velocity, and variety — but overlook one critical dimension: representativeness. In this talk, we will explore the concept of data representativeness, why it matters for building reliable models, and how biased or incomplete datasets can lead to imperfect insights and systemic failures. Are we, as societies, also "training" our systems on biased observations?