News

Master thesis topics 2026-2027 available

DECIDE research at AAMAS

We are pleased to announce that the paper “Interactionless Inverse Reinforcement Learning: A Data-Centric Framework for Durable Alignment” by Elias Malomgré and Pieter Simoens has been accepted for the Blue Sky Ideas track at the 2026 International Conference on Autonomous Agents and Multiagent Systems (AAMAS). AAMAS is the premier conference on multi-agent systems with an A* ranking. The Blue Sky special track (acceptance rate 13%) is intended to present visionary ideas and long-term challenges.

Rethinking AI Safety: Beyond “Alignment Waste”

As AI systems become more complex, ensuring they behave safely and according to human values (a process called “alignment”) is more critical than ever. However, current methods often create what the authors call “Alignment Waste”. Currently, when a new AI model is developed, the safety training from previous models is often lost or cannot be easily reused, making safety an expensive and disposable part of the development process. In this paper, we propose a new framework called Interactionless Inverse Reinforcement Learning (IIRL). This approach decouples safety rules from the specific AI model, creating a “reward model” that is:

• Model-agnostic: it can be applied to different AI systems without starting from scratch.
• Inspectable and Editable: Human experts can clearly see and refine the safety constraints.
• Durable: It transforms AI safety into a permanent engineering asset rather than a one-time fix.

The full article is available here.

GeoAI for disaster response workshop

On June 3rd, 2026, DECIDE will co-organize the GeoAI for disaster response workshop at the GeoAI conference in Ghent. This workshop will explore recent advances in machine learning, remote sensing, and spatial analytics for hazard monitoring, damage assessment, and emergency response during man-made and natural disasters. Please refer to the workshop website for more information.

PhD degree awarded

Thorsten Cardoen was awared his PhD degree for his dissertation entitled “From Pixels to Behaviour: Towards Automated Welfare Assessment through Computer Vision”.

Thorsten developed state-of-the-art camera-based monitoring methods tailored for the challaned of broiler pens. He addresses key challenges in occlusion handling, label efficiency, and fine-grained behavioral quantification. He also engaged in an interdisciplinary collaboration with veterinary experts to demonstrate how his technological innovations can yield actionable insights for precision livestock farming.

PhD degree awarded

The automated analysis of audiovisual urban surveillance data can unlock many practical applications. Deploying such a system requires deep learning algorithms that scale from the confined lab development setting to the challenging real-world context.

In his dissertation, Wei-Cheng Wang identified and tackled three main challenges associated with this lab-to-street transfer: privacy-friendly ML, contrastive learning under sparse event data, and training location-specific ML models without having access to labelled data.

Best paper award

Active Inference is a principled account to integrate perception, planning and action into a single process of free energy minimization. At DECIDE, we employ Active Inference to equip intelligent systems with computational models of human behavior to improve interaction and cooperation.

At IWAI 2025, Alvaro Garrido Perez received the best paper award for his work on modelling cognitive effort through active inference. Apart from a certificate, he received the famous Markov Blanket.

You can read his paper here.

PhD degree awarded

As robots enter our world, it’s not enough that they just seem to work. They must behave in ways that we understand and trust.

In his thesis, Mattijs developed methods to learn symbolic representations from human demonstrations, that are able to capture explicit rules (e.g. traffic laws), implicit norms (e.g. social etiquette) as well as task structure (e.g. get cup before making coffee).

You can find this dissertation here. Part of the research is also explained in our research blog.

PhD degree awarded

In his dissertation, Stef presentes a series of algorithms developed for swarms of biomimetic robots to support animal collectives through exploration and protection tasks, using fish as a case study. He developed these algorithms drawing upon the paradigm of swarm intelligence, in which collective behavior emerges form individuals operating based on simple behavioral rules derived from local interactions with their neighbors and the environment.

Combining algorithms for cage formation and guidance, Stef was able to tackle the complex problem of navigating a fish school along a time-varying safe path. A key assumption is the repulsive effect of robots on the fish individuals, similar to how a shepherd dog repulses a flock of sheep. Stef put his algorithms to the test and validated these assumptions in the Biorobotics Lab of prof. Tim Landgraf. The results indicate more nuanced effect, opening exciting new avenues for research.

You can find this dissertation here.