Ciem Cornelissen
Researcher at IDLab, Ghent University - imec
IDLab, Department of Information Technology
Ghent University - imec
Ghent, Belgium
I am a PhD researcher at IDLab, Department of Information Technology at Ghent University - imec, where I specialize in multi-modal self-supervised learning and world models for autonomous perception. My research focuses on developing AI systems that learn rich, unified representations from diverse sensor modalities without relying on large-scale manual annotation.
My academic journey reflects a deliberate progression from fundamental science to applied AI. I completed my Bachelor’s and Master’s in Physics at Ghent University, where I developed strong analytical foundations and an aptitude for dissecting complex problems at their core. During my physics master’s, elective courses in artificial intelligence sparked a passion that led me to pursue an Advanced Master’s in Artificial Intelligence at KU Leuven (2023), where I specialized in deep learning, computer vision, and data science. This trajectory has uniquely positioned me to tackle the multi-modal representation learning challenges central to my PhD research.
My current research centers on multi-modal self-supervised learning and world models for autonomous systems, leveraging transformer-based architectures to fuse heterogeneous sensor streams such as RGB, LiDAR, and thermal data into coherent, predictive representations. I develop frameworks that learn how the world is structured from unlabelled multi-modal observations, enabling downstream tasks like 3D object detection, depth estimation, and semantic segmentation with minimal supervision. A key focus is designing efficient fusion mechanisms and scalable training objectives that work across diverse sensor configurations and deployment scenarios.
Key Research Contributions:
- Developed Le MuMo JEPA, a multi-modal self-supervised framework that learns unified RGB-LiDAR-thermal representations through learnable fusion tokens, achieving strong performance-efficiency trade-offs on Waymo, nuScenes, and FLIR benchmarks (CVPR 2026 URVIS Workshop)
- Developed LISA (Light-Invariant Spectral Autoencoder), a domain-adversarial deep learning framework enabling robust spectral analysis under varying illumination conditions, part of an IoT-enabled robotic perception system
- Created OHSLIC (Online Hyperspectral Simple Linear Iterative Clustering), an efficient algorithm achieving real-time segmentation on resource-constrained edge devices through adaptive, on-device processing
- Co-authored research on computational fairness in adaptive neural networks, introducing resource allocation disparities as a novel dimension of algorithmic fairness and highlighting ethical considerations in efficiency-driven AI
My work has been published in leading venues including the CVPR 2026 URVIS Workshop, IEEE Internet of Things Journal, IEEE/CVF Winter Conference on Applications of Computer Vision (WACV), and Neural Computing and Applications. Beyond research, I serve as a Teaching Assistant for the Reinforcement Learning course at Ghent University (2024-2026) and supervise master’s thesis students on topics spanning multi-modal self-supervised learning, world models, and autonomous perception.
I am driven by a deep fascination with building AI systems that develop internal world models from multi-modal sensory data, enabling autonomous agents to perceive, predict, and act in complex environments. My research philosophy centers on designing self-supervised objectives and fusion architectures that learn generalisable representations, advancing both the theoretical foundations and practical deployment of multi-modal perception systems.
news
| Mar 26, 2026 | New paper accepted at CVPR 2026 URVIS Workshop!  Our work on Le MuMo JEPA — a multi-modal self-supervised framework for learning unified RGB-LiDAR-thermal representations with learnable fusion tokens — has been accepted! |
|---|---|
| Oct 20, 2025 | New blog post published! Read about our illumination-invariant AI system for in-field grape quality mapping.   |
| Oct 15, 2025 | Deep Dive Talk at FAIR Day 2025 |
| Oct 03, 2025 | New paper published in IEEE Internet of Things Journal! Our work on illumination-invariant deep learning for in-field grape yield and quality mapping is now available. |
| Jun 30, 2025 | New publication in Neural Computing and Applications! Our research on computational fairness in adaptive neural networks explores resource allocation disparities across demographic subgroups. |
