Imperial awarded European COST Action Funding

The European Cooperation in Science and Technology (COST) has awarded funding to establish an innovative network

Imperial has been awarded a COST Action Grant to lead a new European research network focused on edge deep learning for particle physics. The European Cooperation in Science and Technology (COST) is a funding body that supports bottom-up initiatives aimed at building interdisciplinary research and innovation networks across Europe. Each COST Action receives four years of funding to facilitate academic collaborations, promote knowledge exchange in science and technology, and foster innovation.

The Edge Deep Learning for Particle Physics (EPIGRAPHY) network will be led by Dr Benedikt Maier, Eric and Wendy Schmidt AI in Science Postdoctoral Fellow at Imperial’s I-X Centre for AI in Science. The network aims to increase research capacity in deep learning applications in particle physics, with a particular focus on advancing real-time decision-making using artificial intelligence on edge devices in experiments at the Large Hadron Collider (LHC).

AI for Particle Physics

Currently, the volume of collision data produced at the LHC every second far exceeds global storage capabilities, meaning only a small fraction can be retained. To address this problem, the LHC uses a real-time triggering system that determines which collision events to store and which to discard. However, this solution presents a significant computational challenge. To improve real-time decision-making, researchers are working on deploying deep learning algorithms directly on edge devices installed as close to the particle detectors as possible. These AI models will be deployed on hardware accelerators such as FPGAs and ASICs, which are embedded within or close to the detectors of the LHC experiments. However, this environment imposes strict constraints on latency and computational resources, requiring the development of compressed deep learning models that still deliver high predictive accuracy. By applying advanced compression techniques, the ultimate goal is to develop and deploy highly efficient on-edge AI solutions that enable faster and smart event selection at the earliest stage of data acquisition.

The EPIGRAPHY network will bring together researchers from across Europe to work on new deep learning models for use in next-generation experiments at the LHC, particularly during its High Luminosity phase, which begins in 2030. While the primary focus of the initiative is to advance real-time, on-edge AI inference for high-energy physics, the solutions developed will also have broader applications in other domains that rely on real-time AI decision-making on edge devices, including medical technologies and satellite systems. The initiative builds on Dr Benedikt Maier’s postdoctoral project, supported by Schmidt Sciences, which focuses on developing new AI strategies to analyse the collisions at the LHC.

The EPIGRAPHY initiative has been awarded a total of €400,000 over four years. This funding will support a range of activities, including hackathons, data challenges, dataset hosting, workshops, and journal publications. The network includes twenty eighter members, including early career researchers and senior academics, from thirteen countries: Croatia, Cyprus, Estonia, Germany, Greece, Israel, Italy, Poland, Portugal, Serbia, Sweden, Switzerland, Turkey, and the United Kingdom. The network is supported by CERN (Switzerland).

In recent years, the topic of fast, on-edge AI inference has gained significant attention, with numerous research networks emerging, especially in the United States. The EPIGRAPHY network aims to build comparable research capacity in Europe. The initiative also continues the legacy of the EU-funded SMARTHEP project (Synergies Between Machine Learning, Real Time Analysis and Hybrid Architectures for Efficient Event Processing and Decision Making), continuing its mission to advance the use of AI in scientific event processing and decision-making.

Commenting on the importance of the funding for driving innovation in high-energy physics, Dr Benedikt Maier said: “This COST action will help us build momentum and get organized in the European research landscape towards deploying the most powerful algorithms in the most challenging environments of future particle physics detectors”.

New members are now able to join the network, helping to build research expertise across Europe. The network’s kick-off meeting is scheduled to take place in Autumn 2025.