The QUVA Lab centers its research on three core themes: deep vision and video understanding, foundation and adaptive models, and efficient, hardware-aware machine learning.
To drive these advancements, the lab leverages key technical AI components, focusing primarily on:
With a commitment socially aware, and responsible AI systems.
The QUVA Lab is dedicated to bridging the gap between cutting-edge fundamental machine learning research and real-world technology applications.
The lab’s mission and vision are to build the foundational theories, computational models, and algorithmic capabilities necessary to advance the next generation of deep vision and machine learning systems, focusing on core challenges such as computer vision, efficient learning representations, causality, and quantum machine learning.
The impact of the lab lies in driving thought leadership and technical breakthroughs, achieved by transferring fundamental academic research directly into industry applications alongside industrial partners, nurturing top scientific talent, and strengthening the broader AI ecosystem in Amsterdam and internationally.
Adaptable foundation models – Foundation models have established themselves as a revolutionary class of general-purpose AI models that provide impressive abilities to generate text, images, videos, and more. In this project, we study, develop, and evaluate new adaptive learning schemes throughout the foundation model lifecycle, covering pre-training, adaptation, and deployment.
Unsupervised learning for source compression – Learned compression has seen recent advantages relative to traditional compression codecs. In this project, we will explore new methods for lossless and lossy compression with unsupervised learning.
Federated learning – Future of machine learning will see data distributed across multiple devices. This project studies effective model learning when data is distributed and communication bandwidth between devices is limited.
Efficient video representation learning – Despite the enormous advances of image representation learning, video representation learning has not been well explored yet because of the higher computational complexity and the space-time dynamics shaping the content of a video. In this research, we try to capture the content of a video more efficiently in different aspects such as data and computational efficiency.
Video action recognition – Focus on video understanding with the goal of alleviating the dependency on labels. We develop methods that leveraging few-shot, weakly-supervised, and unsupervised learning signals.
Hardware-aware learning – Develop novel approaches for hardware-aware learning, focusing on actual hardware constraints, and work towards a unified framework for scaling and improving noisy and low-precision computing.
Generalizable video representation learning – Aims to develop self-supervised methods that obtain generalizable video representations and solve novel tasks for which the usage of multiple modalities and is a necessity, such as video scene understanding.
Geometric deep learning – Study how to encode the geometry of a problem into neural-network architectures to achieve improved data efficiency and generalization. A particular focus will be given to 3D data and the task of 3D reconstruction, where the global 3D structure is only accessible through a number of 2D observations.
PIN: Positional Insert Unlocks Object Localisation Abilities in VLMs Journal Article
In: CVPR, 2024.
How to Train Neural Field Representations: A Comprehensive Study and Benchmark Proceedings Article
In: CVPR, 2024.
Protect Your Score: Contact Tracing With Differential Privacy Guarantees Proceedings Article
In: Association for the Advancement of Artificial Intelligence, AAAI, 2024.
Towards the Reusability and Compositionality of Causal Representations Proceedings Article
In: 3rd Causal Learning and Reasoning, 2024.
PDE-Refiner: Achieving Accurate Long Rollouts with Temporal Neural PDE Solvers Proceedings Article
In: Thirty-seventh Conference on Neural Information Processing Systems, 2023.
Rotating Features for Object Discovery Proceedings Article
In: Thirty-seventh Conference on Neural Information Processing Systems, 2023.
Time Does Tell: Self-Supervised Time-Tuning of Dense Image Representations Journal Article
In: ICCV, 2023.
Qualcomm is a research and product development organisation in Amsterdam. It is the world’s leading wireless technology innovator and the driving force behind the development, launch, and expansion of 5G.
University of Amsterdam (UvA) is the Netherlands’ largest university, offering the widest range of academic programs.
