The ever wider use of ICT in our society is reflected in the growing complexity of ICT systems and probably, the growing number of cyber criminals. These growing numbers impact the risk of cyber criminality adversely. Risk is an important concept in our research, it is the average impact of a given malicious interaction with an ICT infrastructure. Basically our research goal is to obtain the knowledge to create ICT systems that model their state (situation), discover by observations and reasoning if and how an attack is developing and calculate the associated risks.

In this project, we propose the development of the first large-scale open-source vision-language model from the Netherlands. The model is pretrained at scale using a novel objective that addresses hallucination issues by extending traditional pretraining to include dual-modality predictions of both textual and visual tokens. The architecture combines state-of-the-art components for language processing and vision encoders. Leveraging modern datasets like Cambrian and cutting-edge architectures such as LLaVA Next, we conduct large-scale pretraining on AMD LUMI supercomputing nodes, optimized for computational efficiency and scalability. This work aims to significantly contribute to the vision-language research community by advancing robust multimodal reasoning capabilities.

The PAUL (Playful Active Urban Living) project has investigated how a physical activity (PA) app can be developed that better anticipates the context of the user. This context may refer to the user’s location but also to how PA fits into the user’s activity pattern. Important conclusions are: 1. The exact implementation and design of strategies in the app is crucial for their effectiveness; 2. Feasibility studies deliver much useful feedback about the use of the app; 3. Context aware messages can make PA apps more attractive and effective, but their timing is crucial.

Our planet is changing rapidly. To understand and forecast how ecosystems are affected by global change, ecology should become a predictive science. We will build a unique virtual research environment that will facilitate this transformation, capitalizing on recent advances in Big Data science. This will enable ecologists to link scattered long-term data on plants, animals, and the environment; share methods for data analysis, modelling, and simulation; and build digital replicas of entire ecosystems (“Digital Twins”). This will transform our ability to understand and predict how ecosystems will respond under different scenarios and mitigation measures, fostering scientific breakthroughs and societal impact.

Current AI systems often make decisions in a way that people don’t understand. This potentially creates mistrust, unfairness and brittleness of these systems to changes. A solution is to build systems that learn and use understandable concepts. This requires annotators to label large amounts of data, which is tedious and expensive. Worse, even with all these annotations, AI models can still learn the wrong concepts! The researchers will develop new methods that learn understandable concepts correctly with a high probability from few annotations. They will also use available knowledge on the interactions between the concepts and information on past decisions.