Exploring the role of AI in gravitational-wave science

From 5 to 8 May 2026, CERN hosted the AI for Gravitational Waves Workshop, bringing together researchers working at the intersection of artificial intelligence, machine learning and gravitational-wave science.

The 4 day workshop, with the support of NGT, gathered more than 80 participants both in person and online, exploring how AI is helping reshape the way gravitational-wave signals are detected, simulated, interpreted and processed in real time. Topics varied from ranging from low-latency detection and parameter inference to fast waveform generation, uncertainty quantification, streaming machine-learning pipelines, detector monitoring and predictive maintenance.

But before diving into the technical programme… what are actually gravitational waves? Let’s hear from Eric Moreno (MIT), one of the organziers of the events.

As Eric very well explains, gravitational waves are tiny disturbances in spacetime produced by some of the most energetic events in the universe, such as the collision of black holes or neutron stars. Detecting them requires extremely sensitive instruments, but also increasingly sophisticated computing tools capable of identifying faint signals hidden in large and complex datasets.

This is where AI and machine learning are becoming increasingly important. As current observatories prepare for future observing runs, and as next-generation experiments such as the Einstein Telescope and LISA move closer, the gravitational-wave community is facing a familiar challenge: how to process more data, faster, while preserving scientific accuracy and reliability.

The workshop addressed this challenge from several angles. Sessions covered machine-learning methods for gravitational-wave data analysis, AI-based approaches to simulation and waveform modelling, real-time inference pipelines, detector operations, and the use of accelerated computing architectures such as GPUs and FPGAs.

On Thursday morning, a dedicated Synergy with CERN and NGT session highlighted areas where CERN expertise in AI, real-time data processing and computing infrastructure can connect with the needs of gravitational-wave science.

• Sioni Paris Summers opened the morning with “AI on edge, underground and in space”, presenting examples of how AI can be deployed close to the data source, whether in particle physics experiments, underground infrastructures or space-based systems.
• Fernando Varela Rodriguez followed with “Agentic AI for infrastructure control”, discussing how AI agents could support the operation and control of complex scientific infrastructures.
• Ricardo Rocha then presented “The CERN AI infrastructure”, giving an overview of the computing resources and services available at CERN to support AI-driven research.
• The session concluded with Sabrina Giorgetti’s talk, “Rethinking the HL-LHC CMS Real-Time data processing with NGT”, which connected the discussion directly to the Next Generation Triggers project. Her presentation showed how NGT is working to rethink real-time data processing for the High-Luminosity LHC, developing new approaches to help experiments such as CMS select and process collision data more efficiently.

The workshop also included a broad poster session and a rich programme of presentations across its different themes, with many participants sharing their work on AI applications for gravitational-wave science. Among them, two contributions were recognised at the end of the event. The best poster award went to Ludovica Carbone from the University of Milano-Bicocca for “Pinpointing PTA Single Sources: Sequential SBI for Sky Localization”, which explored how Sequential Simulation-Based Inference can improve the localisation of continuous gravitational waves from supermassive black hole binaries in Pulsar Timing Array data. The best presentation award went to Lucia Papalini from the University of Pisa and INFN-Pisa for “Overlapping signals in 3G detectors: an approach based on Transformers”, presenting a deep-learning approach based on Transformers and Normalizing Flows to address overlapping signals in future third-generation gravitational-wave detectors.