2026 CERN openlab Technical Workshop

4 Mar 2026, 09:00 → 5 Mar 2026, 20:30 Europe/Zurich

500/1-001 - Main Auditorium (CERN)

Antonio Nappi (CERN), Eric Wulff (CERN), Luca Atzori (CERN), Luca Mascetti (CERN), Maria Girone (CERN), Thomas Owen James (CERN)

2026 CERN openlab Technical Workshop

CERN openlab website CERN openlab on LinkedIn

At the 2026 CERN openlab Technical Workshop we will review the R&D projects carried out during the last year and discuss future plans.

The event will feature technical talks, networking sessions over coffee and lunch, and a technology track dedicated to our industrial partners. 

The event will take place in person at CERN between 4-5 March 2026, and it will provide a perfect opportunity for our industrial partners to meet with the students and fellows working on common projects.

In-person attendance is reserved for the CERN community, researchers, research institutions, and industry partners working, or looking to collaborate, with CERN openlab. Registration is mandatory and registrations will be reviewed one by one.

The workshop will be available via webcast for all audiences, ensuring broad access to the projects updates, keynotes and discussions. Registration is not needed.

Find the webcast here: https://live.cern/event/i1587839

We hope to see many of you taking part in this event! 

Parking Places.png

Wednesday 4 March

09:00 Registration

Welcome & Workshop Goals

Convener: Dr Maria Girone (CERN)

1 Welcome & Workshop Goals

Speaker: Dr Maria Girone (CERN)

2026 TW Welcome 19.12.55.pdf

2026 TW Welcome 19.12.55.pptx

Recording

Infrastructures and Techniques for AI and HPC

Convener: Dr Maria Girone (CERN)

2 Openlab for CERN R&D: the Next Generation Triggers project

CERN's Next Generation Triggers (NGT) project is a large-scale, multi-year R&D project, fusing research from ATLAS, CMS and CERN's Experimental Physics, Information Technology, and Theory Departments. Despite the project's dedicated funding for hardware, the collaboration with Openlab turned out to be essential in several aspects. This presentation will highlight the role of large-scale R&D projects at CERN and how NGT, Openlab and the wider CERN community will benefit from such collaboration.

Speaker: Axel Naumann (CERN)

Openlab-NGT-Naumann-2026-03-04.pdf

Recording

3 Commercial Off the Shelf Data Acquisition for the HiLumi era

The remarkable progress in network and compute technologies, driven today mostly by AI, accelerates a trend to “commoditise” the read-out and on-detector processing of large scientific instruments. I will illustrate these trends using the planned upgrades / evolutions of the readout-systems of the LHC experiments, in particular the phase 2 upgrades of LHCb. There will also be some thoughts on possible lessons for smaller experiments non-HEP going forward.

Speaker: Niko Neufeld (CERN)

COTS_DAQ_Systems.pdf

Recording

4 Anomaly Detection for Ultra Low Latency Event Selection at the LHC (ATLAS)

The LHC is a particle accelerator which can provide up to 40 MHz of proton-proton collisions to its various experiments. In ATLAS, this data is filtered in real time down to 100 kHz in order to meet the readout constraints. This task is performed by the L1 Trigger, which uses FPGA-based hardware to select the most relevant events within a few microseconds. Traditionally, L1 trigger algorithms have been designed to target a set of interesting, predefined Physics signatures. In contrast, anomaly detection triggers aim to flag any events that deviate from expected behaviour, offering sensitivity to rare or unforeseen phenomena. This presentation summarises the latest developments from Run 3, which will also be a valuable input for the design of anomaly detection triggers in the High Luminosity LHC.

Speaker: Paula Martinez Suarez (CERN)

PMS_CERNOpenlabWorkshop_20260304.pdf

Recording

10:40 Coffee break

Infrastructures and Techniques for AI and HPC

Convener: Thomas Owen James (CERN)

5 Investigation of Anomaly Detection algorithms for filtering events with microseconds latency in the ATLAS hardware-level trigger

This project aims to evaluate the robustness of a candidate data-driven autoencoder-based anomaly detection algorithm for use at the ATLAS hardware trigger, which performs real-time event selection to save only the events deemed most interesting. This work tests the sensitivity of the algorithm across a variety of physics processes and explores new methods for increasing sensitivity to anomalous events occurring at low energy or low multiplicity, which are more easily obscured by background.

Speaker: Tara Tahseen (UCL)

Recording

TaraTahseen-CERN-OpenLab-Technical-Workshop_v3.pdf

TaraTahseen-CERN-OpenLab-Technical-Workshop_v3.pptx

6 A transformer on FPGA for low latency applications in particle physics and beyond

To identify physics phenomena that are difficult to detect by standard triggers, real-time modern AI triggers are needed. One of the most promising novel trigger classes is anomaly detection triggers, capable of flagging rare events without relying on a full theoretical understanding of the underlying process. By leveraging modern, heterogeneous accelerator platforms, like the AMD Versal Adaptive Compute Acceleration Platform, which combines classic Field Programmable Gate Array (FPGA) logic with higher-level AI Engines, we aim to achieve ultra-low-latency inference at a high throughput, necessary for operations in the CMS trigger and scouting system. We present a machine learning architecture based on a Transformer model, specifically adapted for low-latency inference. As the Transformer architecture has shown great success in interpreting complex relationships across many areas of research, we believe that a fast, flexible, high-throughput Transformer model, implemented 'on the edge', has wider-reaching applications across technology and industry.

Speaker: Elias Leutgeb (CERN)

Recording

Transformer_FPGA_el.pdf

7 Full event interpretation with MLPF reconstruction in the CMS detector

The particle-flow (PF) algorithm reconstructs a global description of each collision by producing a comprehensive list of final-state particles. It is central to event reconstruction in the CMS experiment at the CERN LHC, and has been a focus of developments in light of planned high-luminosity running conditions with increased pileup and detector granularity. The existing PF implementation relies on several physics-motivated heuristics and assumptions that can be replaced by machine learning (ML) models trained directly on simulated data. A state-of-the-art, ML-based PF (MLPF) reconstruction algorithm implemented within the CMS software framework is presented. The MLPF algorithm performs a learnable, differentiable full-event reconstruction on GPUs, generalizes across detector conditions and collision energies, and replaces multiple traditional reconstruction steps with a single unified model. Physics performance comparable to standard PF reconstruction is achieved in both simulation and data, with improved jet energy resolution and inference time. In simulated top quark-antiquark events under LHC Run 3 (2023–2024) conditions, the jet energy resolution is improved by 10–20% for jets with transverse momentum between 30 and 100 GeV. Runtime performance is evaluated using simulated QCD multijet events with pileup corresponding to 55–75 interactions. For these events, a median runtime of 20 ms per event is achieved on an Nvidia L4 GPU, with better scaling with event size than the standard CMS particle-flow reconstruction, which processes the same events in approximately 110 ms per event.

Speaker: Eric Wulff (CERN)

Eric_Wulff_openlab_technical_workshop_2026_MLPF_in_CMS.pdf

Recording

8 NVidia Holoscan-based R&D for the trigger

High-energy physics trigger systems are traditionally implemented in a two-stage architecture. In the first stage, custom hardware boards—typically FPGA-based—perform fast, coarse reconstruction to identify interactions of interest and forward the selected events to the next stage. There, reconstruction is carried out in greater detail using commercial computing technologies such as CPUs and GPUs. Because of limited resources, high bandwidths, and strict latency constraints, the hardware trigger cannot apply selections as sophisticated as those in the second stage. Technologies such as NVIDIA Holoscan aim to address this limitation by enabling high-bandwidth data streams to be transferred directly into GPU memory, bypassing delays introduced by a CPU host. This talk presents the R&D activities within the ATLAS experiment evaluating the Holoscan technology, initially in the context of data scouting setups, and explores the feasibility of GPU-based trigger and data-acquisition systems for future experiments.

Speaker: Ioannis Xiotidis (CERN)

AGHOST_OpenLabWorkshop.pdf

Recording

9 From Specialized Algorithms to Foundation Models

Speaker: Eric Wulff (CERN)

Eric_Wulff_openlab_technical_workshop_2026_Simons_v2.pdf

Recording

12:40 Lunch + Group Photo

Heterogeneous Computing, Platforms and HPC

Convener: Thomas Owen James (CERN)

10 Keynote: The HPC challenge of radio astronomy

Radio astronomers are engaged in an ambitious new project to detect faster, fainter, and more distant astrophysical phenomena using thousands of individual radio receivers linked through interferometry. The expected deluge of data (up to 300 PB per year) poses a significant computational challenge that requires rethinking and redesigning the state-of-the-art data analysis pipelines. In this talk we present an overview of the Square Kilometer Array Observatory, the HPC challenges, and solutions using physics-informed machine learning.

Speaker: Dr Emma Elizabeth Tolley (EPFL)

HPCRadio_CERNopenlab.pdf

Recording

11 Powering AI Inference from Edge to Cloud for Modern Business

Discover how solutions with Intel® Xeon® 6 and Intel® Core™ Ultra processors revolutionize AI inference across your entire infrastructure. This presentation explores the compelling business case for modernizing compute environments, featuring breakthrough TCO improvements, enhanced sustainability metrics, and seamless AI workload deployment from edge to cloud. Learn how Intel’s tools and software ecosystem support a common software foundation for both HPC and AI workloads, enabling consistency across diverse use cases. Built to operate across heterogeneous systems, this software stack integrates with modern orchestration and scheduling environments to support scalable deployment and efficient resource utilization. It helps solve some of the challenges that large‑scale infrastructures and data‑intensive environments are currently facing, particularly around specialized data collection requirements and low‑latency data handling, which are increasingly driving the adoption of AI‑based filtering and inference. Together, hardware and software form a cohesive platform approach that helps organizations evolve their infrastructure while maintaining flexibility over time.

Speakers: Jean-Laurent Philippe (Intel), Nisha Patel (Intel)

CERN Openlab 4 Mar 2026 - Nisha-JLP.pdf

Recording

12 Overview of the LHCb Allen project

The Allen project is simultaneously a framework for portable parallel computing, optimised for applications which require extreme throughputs, and the concrete implementation of LHCb's GPU-based first-level trigger system deployed in LHC Run 3. This talk will cover the genesis of Allen, its performance as a production application in Run 3, and discuss current efforts to refactor the experiment-independent parts of Allen (AllenCore) into a general-purpose high-throughput cross-architecture processing framework which can work together with other cross-experiment frameworks and libraries such as Gaudi and ACTS/Traccc.

Speaker: Vava Gligorov (Centre National de la Recherche Scientifique (FR))

Allen_OpenLab_2026.pdf

Recording

13 An overview of AI projects at CERN with applications in industry

Artificial Intelligence is becoming a central pillar of scientific discovery and technological innovation at CERN. This talk will provide an overview of AI activities across CERN, with a focus on Fast Machine Learning and its use in real-time environments at the LHC experiments. I will discuss how ultra-low latency, resource-constrained AI solutions are enabling intelligent triggering and data processing close to the data source on detectors. Beyond high-energy physics, I will highlight collaborations with industry, where techniques developed for extreme scientific environments are adapted to edge-computing scenarios. Particular emphasis will be placed on ultra-fast, low-power, and TinyML approaches, demonstrating how innovations driven by fundamental research can contribute to efficient, deployable AI systems across sectors.

Speaker: Sioni Paris Summers (CERN)

openlab_workshop_2026_sps.pdf

Recording

14 Porting and Performance Assessment of a Gravitational N-body code on the RISC-V-Based Tenstorrent Wormhole accelerator

The open, modular, and royalty-free RISC-V Instruction Set Architecture (ISA) is attracting growing interest within the High Performance Computing (HPC) community as a promising alternative to proprietary architectures. Among RISC-V–based solutions, Tenstorrent accelerators, originally developed with Machine Learning and Artificial Intelligence workloads in mind, are also appealing for scientific HPC applications due to their ability to decouple data movement from computation.
To investigate this potential, we developed an N-body code for astrophysical simulations and offloaded its most computationally intensive kernel to Tenstorrent RISC-V–based accelerators using the TT-Metalium programming interface. We evaluated the performance of the resulting implementation in terms of execution time and energy consumption by running a representative simulation on the n300 Wormhole card, and compared the results against both a highly optimized CPU version and a CUDA-accelerated implementation.
Our results show that, compared to the CPU baseline, the TT-Metalium implementation achieves more than a twofold speedup while reducing energy consumption by approximately 50%. In comparison with NVIDIA GPUs, the CUDA implementation currently delivers higher absolute performance. Nevertheless, despite this gap with respect to state-of-the-art accelerators, the Tenstorrent card yields encouraging results for this class of workloads, suggesting that further hardware and software developments may enhance its competitiveness.
Finally, to the best of our knowledge, this work represents the first exploration of strategies for scaling a scientific application across multiple Tenstorrent accelerators, with experiments conducted on configurations of up to four devices, highlighting the feasibility of multi-accelerator deployments for HPC workloads on this emerging platform.

Speaker: Elisabetta Boella (E4 Computer Engineering SpA)

2026_02_04-E4-Computer-Engineering-CERN.pdf

Recording

15:45 Coffee break

Heterogeneous Computing, Platforms and HPC

Convener: Luca Atzori (CERN)

15 NVIDIA Data Center Platform Update

This talk will give an overview of NVIDIA’s data center platform, including updates on GPU, CPU and Networking technologies.

Speaker: Sebastian Kalcher (NVIDIA)

NVIDIA Platform Update-openlab.pdf

Recording

16 Heterogeneous Architectures Testbed (HAT)

This project aims to provide a diverse hardware portfolio for comprehensive technology testing. Focused on assessing the efficacy of various architectures, this initiative aims to provide valuable insights into the practical utility of emerging technologies. By subjecting a spectrum of hardware configurations to real-world applications, the project seeks to establish benchmarks that guide the adoption of the most effective and efficient technologies.

Speakers: Albane Carcenac (CERN), Jessy Sobreiro (CERN)

openlab_HAT_BEAM_2026.pdf

openlab_HAT_BEAM_2026.pptx

Recording

17 HEPScore Suite Enhancements: Power Metrics and GPU Integration

The HEPiX Benchmarking Working Group develops and maintains benchmarking tools to quantify computing resources across the Worldwide LHC Computing Grid (WLCG). Since the adoption of HEPScore23 as the official WLCG benchmark in April 2023, the WG has been actively enhancing the benchmark suite based on community feedback and emerging requirements.

Currently, the WG is advancing two key initiatives.

First, the development of instrumentation modules that capture comprehensive server utilization metrics during HEPScore execution, including CPU load, frequency scaling, I/O patterns, and power consumption. These metrics enable detailed analysis of power efficiency and performance characteristics across different hardware configurations.

Second, the integration of GPU workloads into the benchmark catalog, expanding HEPScore’s applicability to heterogeneous computing environments and modern accelerator-based systems.

In this report, we provide an overview of the WG’s technical progress, implementation challenges, and preliminary results from these developments.

Speakers: Natalia Diana Szczepanek (CERN), Robin Hofsaess (CERN)

HEPScore_Enhancements_Openlab_Workshop_20260304.pdf

Recording

18 Towards unified full-stack performance analysis and automated computer system design with Adaptyst

Slow performance is often a major blocker of new visionary applications in scientific computing and related fields, regardless of whether it is embedded or distributed computing. This issue is becoming more and more challenging to tackle as it is no longer enough to do only algorithmic optimisations, only hardware optimisations, or only (operating) system optimisations: all of them need to be considered together.

Architecting full-stack computer systems customised for a use case comes to the rescue, namely software-system-hardware co-design. However, doing this manually per use case is cumbersome as the search space of possible solutions is vast, the number of different programming models is substantial, and experts from various disciplines need to be involved. Moreover, performance analysis tools often used here are fragmented, with state-of-the-art programs tending to be proprietary and not compatible with each other.

This is why automated full-stack system design is promising, but the existing solutions are few and far between and do not scale. Adaptyst is an open-source project at CERN aiming to solve this problem. It is meant to be a comprehensive architecture-agnostic tool which:

• unifies performance analysis across the entire software-hardware stack by calling state-of-the-art software and APIs under the hood with any remaining gaps bridged by Adaptyst (so that performance can be inspected both macro- and microscopically regardless of the workflow and platform type)
• suggests automatically the best solutions of workflow performance bottlenecks in terms of one or more of: software optimisations, hardware choices and/or customisations, and (operating) system design
• scales easily from embedded to high-performance/distributed computing and allows adding support for new software/system/hardware components seamlessly by anyone thanks to the modular design

The tool is in the early phase of development and concentrating on profiling at the moment. Given that Adaptyst has broad application potential and we want it to be for everyone’s benefit, we are building an open-source community around the project and looking for industrial and academic collaborations.

This talk is an invitation to join us: we will tell you in detail what Adaptyst is and how you can get involved.

Speaker: Maksymilian Graczyk (CERN)

Adaptyst openlab talk.pdf

Recording

17:35 Networking Cocktail & 25th Anniversary Celebration

Thursday 5 March

Storage

Convener: Luca Mascetti (CERN)

19 Data and Computing Challenges in Experiment Upgrades: the ALICE case

The ALICE experiment was originally designed as a relatively low-rate experiment, mainly due to constraints imposed by the Time Projection Chamber (TPC) readout system utilising
MWPCs. However, following hardware upgrades during the LHC LS2, which include a GEM-based continuous readout for the TPC, ALICE is now capable of operating at a peak Pb-Pb collision rate of 50 kHz. Notably, all events are processed, reconstructed, compressed, and written to permanent storage without relying on any selective triggers.
The talk will describe the ALICE online data processing system, specifically designed to handle the substantial raw data rate of ~3.5 TB/s generated by the detectors.

Speaker: Andreas Morsch (CERN)

OL Seminar 2026.pdf

Recording

20 Building Scalable Flash Storage with Pure EXA

High-performance storage for scientific computing now demands exabytes of data readable and writeable at terabytes per second. Flash storage is poised to replace disk, since it is faster, more energy-efficient, and more space-efficient than an equivalent amount of disk. This talk will describe Pure Storage's //EXA architecture, which provides exabyte-scale flash-based storage for HPC and AI using a high-performance metadata server and a large number of data nodes. The talk will show how a flash-based design can reduce rack space and energy consumption by an order of magnitude over disk-based systems while providing comparable bandwidth and better reliability.

Speaker: Ethan Miller (Pure Storage)

Pure Scalability + EXA CERN 2026.pdf

Recording

10:40 Coffee break

21 Next-Generation Exascale Flash Storage

This project aims to evaluate next-generation, high-density flash-based storage technologies through a strategic CERN openlab – Pure Storage collaboration. By combining CERN’s exascale operational expertise with Pure Storage’s high-efficiency DirectFlash platform, the initiative will assess performance, scalability, energy efficiency, cost, and reliability. The overall goal is to determine whether such technologies can sustainably and cost-effectively support future scientific data volumes at exabyte scale.

Speaker: Ruhi Choudhury (CERN)

2026-03-05_Next-gen_Exascale_Flash_v3.pdf

2026-03-05_Next-gen_Exascale_Flash_v3.pptx

Recording

Storage

Convener: Luca Mascetti (CERN)

22 Towards Archival Storage Using QLC Flash

This talk discusses an alternate approach to archival storage: low-cost, high-capacity flash storage. Long-term archival storage is typically implemented using disk or nearline media such as tape and optical. While this approach is relatively inexpensive, it limits access to archival data behind a slow interface with limited bandwidth. In contrast, high-density QLC flash storage shipping today can fit half an exabyte into a single rack, with increases in density forecast to exceed an exabyte per rack in less than 18 months. QLC flash has low power consumption, much higher read bandwidth, and longer durability than existing archival storage, More excitingly, archival QLC flash enables new ways of interacting with archival storage via "background" experiments that piggyback on periodic data scrubbing to leverage the massive bandwidth available at low cost and power from flash storage. Using this model, researchers can run many more experiments on their archived data at relatively low cost compared to today's approach of fetching archival data to "fast" storage and analyzing it.

Speaker: Ethan Miller (Pure Storage)

Archival flash + QLC CERN Apr 2026.pdf

Recording

23 Ceph Scaling Strategies for Machine Learning Workloads

Ceph is a software-defined solution designed to manage large-scale data stores. It ensures high availability and concurrent access while providing a POSIX-compliant file system (CephFS), making it an attractive option for serving distributed machine learning workflows. This project focuses on optimising CephFS resource usage to improve overall storage performance and reduce the bottleneck of data access in such use-cases. We will present our testbed and benchmarks of various CephFS configurations to examine possible optimizations from cluster-administrator and user perspectives.

Speaker: Radomir Michal Wasowski (CERN)

Openlab2026.pptx

Openlab2026.pptx.pdf

Recording

24 Cerabyte - Building the Ecosystem

Often significant public money has been spent to run research projects and collect data, thus propagating this data and keeping it available for coming generations of researchers is an obligation we have to fulfill. Furthermore, it is likely that future researchers will be able to gain additional insights from this data. The challenge is how to afford the preservation of this data in a sustainable way while keeping it accessible. Cerabyte is a technology to permanently preserve huge sets of digital data, cost-efficient and sustainable, with fast access when needed. We provide an update on the ecosystem and technology progress to date including the current status and remaining challenges as well as our collaboration with openlab and an outlook on planned activities for 2026.

Speaker: Steffen Hellmold (Cerabyte)

CERN openlab workshop 2026 - Cerabyte vFinal with video.pdf

CERN openlab workshop 2026 - Cerabyte vFinal with video.pptx

Recording

25 Evaluation of Cerabyte technology - summary of activities in 2025

We will summarize various activities that we have done in 2025 related to the evaluation of the Cerabyte data storage technology.

Speaker: Vladimir Bahyl (CERN)

Cerabyte evaluation - 2025 activities.pdf

Cerabyte evaluation - 2025 activities.pptx

Recording

12:45 Lunch

Computing Architectures and Software Engineering

Convener: Eric Wulff (CERN)

26 BioDynaMo: Recent Developments and Future Directions

Agent-based modeling provides a powerful framework to study complex biological systems across multiple spatial and temporal scales. As research questions in areas such as cancer immunotherapy and infectious disease modeling grow in complexity, simulation platforms must combine biological realism with computational scalability while remaining flexible enough to support evolving scientific needs.

BioDynaMo is an open-source, high-performance agent-based simulation platform that empowers researchers to generate, execute, and visualize large-scale biological simulations. Designed with a modular architecture and leveraging expertise in large-scale computing at CERN, BioDynaMo enables complex simulations while maintaining adaptability for diverse biomedical applications. This talk presents recent developments of the project and outlines future directions aimed at expanding its scientific scope and impact.

Speaker: Stavros Portokalidis (CERN)

BioDynaMo-openlab-workshop-2026.pdf

BioDynaMo-openlab-workshop-2026.pptx

Recording

27 Next Generation Archiver for WinCC OA: Status & Outlook

WinCC OA is a Supervisory Control and Data Acquisition (SCADA) software platform deployed in more than 850 systems at CERN. It includes mission-critical installations such as the electrical grid, cryogenics, vacuum systems, detector control systems, and cooling and ventilation. Such systems continuously generate operational and experimental data whose long-term preservation and accessibility are essential for analysis and safe operation. For instance, variations in detector states across experiments are crucial for physics analysis, while abnormal events or alarms may signal conditions relevant to safe and reliable operation. However, the high volume of generated data presents significant technical challenges. The Next Generation Archiver (NGA) addresses these challenges by providing an efficient interface for storing and accessing data across heterogeneous database systems via pluggable backends. The introduction of a TimescaleDB backend further strengthens support for time-series data generated by WinCC OA. This presentation highlights recent developments in the NGA project, including the integration of TimescaleDB and an evaluation of its performance.

Speaker: Nikita Nekhotyachshiy (CERN)

NextGenerationArchiver2026.pdf

Recording

Infrastructures and Techniques for AI and HPC

Convener: Eric Wulff (CERN)

28 An overview on Virtualized PLCs and Agentic AI for Industrial Control Systems

As preparations for the Future Circular Collider (FCC) advance, new technical and organizational challenges emerge. The FCC's highly distributed geographic nature and complex control systems pose significant obstacles to efficient development and operation. The unprecedented scale of the project, with its vast number of components and interdependent tasks, demands innovative approaches to remain effective.

Introducing new technologies and methodologies is essential to contain construction and operating costs while ensuring reliability. Virtualized PLCs provide an effective solution for managing remote physical components in an easily automated and configurable environment. Concurrently, AI agents can accelerate complex workflows, integrate cross-domain knowledge more effectively, and free experts from repetitive, low-complexity tasks.

This presentation covers two concrete developments in these areas. First, we discuss a test bed deployment of virtualized PLCs within an edge-cloud continuum architecture, demonstrating how this approach enables flexible resource allocation and improved system resilience for distributed control systems. Second, we present the development of custom tools for AI agents specifically designed for industrial control code generation, showing how these specialized capabilities can enhance productivity.

These advancements represent practical steps toward addressing the unique challenges of next-generation particle accelerator projects.

Speaker: Filippo Berto (CERN)

berto - vplc and ai agents.pdf

Recording

Heterogeneous Computing, Platforms and HPC

Convener: Antonio Nappi (CERN)

29 Oracle: Embracing Open Source and AI

An overview of the open-source initiatives we are actively developing and contributing to in collaboration with CERN, highlighting our work on OraOperator and Observability Exporter. It also reflects Oracle's commitment to innovating and contributing to leading open-source AI platforms. The session explores the challenges these projects address, the impact of our contributions, and how open collaboration accelerates innovation across the ecosystem.

Speakers: John Lathouwers (Oracle), Marco Stefanetti (Oracle), Matteo Malvezzi (Oracle)

CERN_Openlab_2026.pdf

Recording

15:10 Coffee break

Heterogeneous Computing, Platforms and HPC

Convener: Antonio Nappi (CERN)

30 Modernization of Oracle REST Data Services (ORDS) for Kubernetes

Oracle REST Data Services (ORDS) is a Java middleware that exposes Oracle Database into RESTful APIs. Instead of connecting through traditional database drivers, it allows applications to interact with the database over HTTP using JSON.

In environments running dozens or hundreds of ORDS instances, maintaining local configuration files becomes complex and error-prone. Recently introduced, the Central Configuration Server is a REST API that serves as a centralized configuration source for ORDS instances, providing a scalable way to deploy, update, and standardize configuration across all instances. By externalizing configuration and enabling stateless deployments, it makes ORDS architectures more resilient, industrialized, and Kubernetes-ready. It also integrates seamlessly with the Oracle Kubernetes Operator for automated lifecycle management.

Speaker: Thomas Saury (CERN)

CERN_Thomas_Saury_openlab_workshop_2026_03.pdf

Recording

31 Carbon-Aware FinOps for On-Premises and Public Cloud

With GPUs pushing the needs for power and cooling, cost optimization and sustainability must go hand in hand. Small, data-driven efficiency gains can yield significant cost and carbon reductions.

This talk presents the work being done around carbon-aware FinOps in the OpenLab/Oracle collaboration, including an early proof-of-concept deployed across multiple infrastructures, from cloud-native environments to the Next Generation Triggers (NGT) cluster. We'll show how combining hardware telemetry with model-based estimation promises smarter optimization from datacenters down to individual workloads.

Speaker: Laura Eve Sarah Llinares (CERN)

OpenlabDays-March2026.pptx.pdf

Recording

Computing Architectures and Software Engineering

Convener: Eric Wulff (CERN)

32 Building a Digital twin of the CMS ECAL

Speakers: Haotian Li (Imperial College London), Shuwen Zheng (Imperial College London)

Building a Digital Twin of the CMS ECAL.pdf

Building a Digital Twin of the CMS ECAL.pptx

Recording

Infrastructures and Techniques for AI and HPC

Convener: Eric Wulff (CERN)

33 Geometric Quantum Machine Learning with Neutral Atoms (Pasqal)

This collaboration focus on neutral atom quantum computing techniques. The research collaboration aims to contribute to advance knowledge and tools that will be strategic to both Parties. The substantial innovation in the field of AI in the past years together with the rapid prototyping of quantum technologies, has enabled the definition of quantum machine learning. It is an active field of research which seeks to take advantage of the capabilities of both quantum computers and machine learning techniques, adapting the latter to the strengths of the former.

Starting from a deep understanding of current classic and quantum implementation of theoretical and computational models for graph neural network, we will focus on scalability, symmetry properties and generalization. In particular, geometric deep learning (GDL) will be a key focus of this work package. The possibility of testing and implementing those model for HEP use cases represent an important test bed for future LHC computing requirements and quantum technologies.

Speaker: Jogi Suda Neto (University of Alabama (US))

CERN Openlab Workshop.pdf

Recording

Closing remarks

34 Closing Remarks

Speaker: Dr Maria Girone (CERN)

Recording