CERN openlab Summer Student Lightning Talks (1/2)

Monday 11 Aug 2025, 13:30 → 16:37 Europe/Zurich

31/3-004 - IT Amphitheatre (CERN)

On Monday 11th and Tuesday 12th of August, the 2025 CERN openlab summer students will present their work at two dedicated public Lighting Talk sessions.

In a 5-minute presentation, each student will introduce the audience to their project, explain the technical challenges they have faced and describe the results of what they have been working on for the past two months.

It will be a great opportunity for the students to showcase the progress they have made so far and for the audience to be informed about various information-technology projects, the solutions that the students have come up with and the potential future challenges they have identified.

Please note 

• Only the students giving a talk need to register for the event
• There are 18 places available on Monday and 19 places on Tuesday
• The event will  be accessible via webcast for an external audience (Please invite your university professors and other students)

Day 2 information: https://indico.cern.ch/event/1543702/ 

Please note that pictures and videos might be taken during the event. The pictures and videos might be used for communication about the event. By joining the lecture, you are agreeing to being featured in these communication actions. 

13:30 → 13:35 Welcome by the CERN openlab team

lighting_talk_welcome.pdf

13:35 → 13:42 Binary Neural Networks

All major CERN experiments along with the accelerator will enter soon in the phase of a big upgrade cycle, called the High-Luminosity LHC, in order to further broaden the physics reach. ATLAS has a plethora of upgrades concerning the HL-LHC era which will equip the detector with many exciting new opportunities. One of the core upgrades in ATLAS concerns the way of reading out the calorimeter sub-detector. In contrast to previous runs ATLAS will be able to read the full granularity of the calorimeter at the level of the hardware trigger system. Having this information available in such a challenging environment provides a unique opportunity to explore Machine Learning ideas on the edge within the context of the Next Generation Trigger project. With the current project we explore the feasibility of implementing Binary Neural Networks (BNNs) in FPGA solutions dealing with high data throughput. BNNs offer the advantage of significantly reduced computational cost and memory usage by operating with binary weights and activations, while still achieving competitive performance. Combining such a novel ML-based technique with the advantages gained with the use of FPGAs at the level of the hardware trigger manifests a unique opportunity to explore cutting edge techniques for data processing. The work involves converting existing CNN-based approaches for tau lepton identification into Binary Neural Network (BNN) counterparts. These models are being evaluated through holistic comparisons in terms of processing power requirements, power estimates, and physics.

Speaker: Maria Mastoreka (Aristotle University of Thessaloniki (AUTH))

Binary_Neural_Networks_for_FPGAs Mastoreka Maria .pdf

13:42 → 13:49 Cold Storage for Science and Beyond: Enabling S3 Glacier on CTA

CERN has a provisioned storage capacity of 1.25 exabytes (EB), of which 99% is used for physics archival data and 1% for backups. Currently, the backup workload is utilizing 20 tape drives. Considering that CERN has a total of 200 tape drives, this means that 1% of the data is consuming 10% of the tape infrastructure. Ideally, the backup system should be using only 2 tape drives, proportional to its data share. This infrastructure imbalance exists because a separate solution is being used for backups, distinct from the one used for physics archival data. This project aims to address the issue by creating a unified endpoint for the three storage classes: S3, Glacier, and CTA.

Speaker: Sarthak Negi

Lightning Talk Sarthak Negi.pdf

Lightning Talk Sarthak Negi.pptx

13:49 → 13:56 Optimising the operation of GPUs to reduce power consumption

Several studies have been conducted to understand the performance of CPUs operating at different clock frequencies. These studies have been also done with HEP related, standardised, workloads, looking at performance and power consumption. For GPUs the clock rate and the operation voltage can be varied. The standard voltage settings ensure a stable operation of all GPUs of the same model. However, most individual GPUs can be operated reliably also at lower voltage settings and therefore also with lower power consumption. Our team has access to a number of HEP related workloads for GPUs. With these workloads a systematic study of the power consumption, performance and stability of several GPUs in our computing centre should be conducted with different voltage and frequency settings. These findings can be expressed as reductions of the CO2 footprint of the workloads. This will be a first step towards a standardised approach for finding the optimal operating point.

Speaker: Sana Babayan Vanestan (Sharif University of Technology)

Sana_Babayan_Vanestan.pdf

13:56 → 14:03 Calorimeter based vertexing

All major CERN experiments along with the accelerator will enter soon in the phase of a big upgrade cycle, called the High-Luminosity LHC, in order to further broaden the physics reach. ATLAS has a plethora of upgrades concerning the HL-LHC era which will equip the detector with many exciting new opportunities. One of the core upgrades in ATLAS concerns the way of reading out the calorimeter sub-detector. In contrast to previous runs ATLAS will be able to read the full granularity of the calorimeter at the level of the hardware trigger system. Having this information available in such a challenging environment provides a unique opportunity to explore Machine Learning ideas on the edge within the context of the Next Generation Trigger project. With the current project we would like to explore the concept of enhancing the vertexing capabilities of ATLAS by using only the calorimeter information for inference. The bulk of the work will be focused on designing and implementing a Symbolic Regression based model which during training will include both the tracking detector information and the calorimeter cells and eventually aim to run inference only with the calorimeter cells. The selected student will be able to work on a novel technique that has not been attempted in the past and will gain valuable insights on developing ML algorithms which will be implemented in cutting edge FPGA solutions.

Speaker: Rafail Athanasios Giannoulakis

CalorimeterBasedVertexing.pdf

14:03 → 14:10 Optimising Multi-Cluster Kubernetes for Active-Active Deployments Across Public and On-Premise Data Centers

Active-Active deployments across multiple data centers are the gold standard for achieving highly available and fault resilient services. However, distributing workloads across many clusters presents significant challenges particularly for stateful deployments. This project will focus on enabling active-active kubernetes deployments across CERN’s data centers by: * Evaluating existing CNI plugin support for cross-cluster networking. * Experimenting with service mesh based solutions to provide inter-cluster traffic routing, load balancing and automated failover. * Investigating current offerings in the CNCF for multi-cluster active-active database deployments. The outcomes of this project will help establish a robust framework for active-active Kubernetes deployments, supporting business continuity and disaster recovery efforts at CERN.

Speaker: Olli Luca Glorioso (OpenLab Summer Intern)

OLLI GLORIOSO PRESENTATION.pptx

14:10 → 14:17 Implementing and benchmarking support for Xilinx AI engines in hls4ml

This project focuses on integrating support for Xilinx AI Engines (AIEs), part of Versal ACAP platform, into the hls4ml framework, enabling the generation of AIE-compatible code for machine learning (ML) inference. The student will map ML operations to AIE capabilities, implement scheduling for parallel execution, and develop a benchmarking suite to evaluate AIE performance in terms of latency, throughput, and energy efficiency compared to existing hls4ml backends. The final deliverables will include a functional AIE backend, documentation, and example workflows, contributing to cutting-edge ML hardware acceleration and expanding hls4ml's ecosystem.

Speaker: Nuno Alexandre Lozano

CERN_OL_Presentation.pdf

CERN_OL_Presentation.pptx

14:17 → 14:24 Energy Analysis for Distributed Machine Learning on HPC with itwinai

As part of CERN’s contribution to the InterTwin project—a European initiative to standardize digital twin development for science—the itwinai team is building a Python toolkit that embeds advanced ML workflows into the InterTwin engine. These workflows often run on distributed HPC systems and demand significant computing resources. This project focuses on improving the usability and sustainability of those workflows by implementing a monitoring feature that tracks GPU, CPU, RAM, disk, and energy consumption. By providing researchers with transparent insights into their resource usage, this work supports more efficient and environmentally conscious deployment of digital twin applications across domains like astronomy and particle physics.

Speaker: Mackenzie Bowal

MackenzieBowal_EnergyAnalysis.pptx

14:24 → 14:31 Scalability Analysis of AI Workloads on HPC with itwinai

Digital twins are increasingly used in scientific research, but ensuring their scalability across high-performance computing (HPC) environments remains a challenge. The interTwin project aims to address this by providing a unified digital twin engine, while interLink offers a mechanism for efficient workload offloading. This project will focus on performing scaling tests of interTwin AI workflows on HPC. The goal is to assess the scalability of both AI workflows—defined using itwinai—and the offloading mechanism provided by interLink. The scaling tests will involve both distributed machine learning (ML) training and hyperparameter optimization across multiple nodes, evaluating key metrics such as computation time, communication overhead, and resource utilization. As a summer student, you will run experiments on HPC systems, analyze scaling performance, and contribute to optimizing the execution of ML workflows. You will gain hands-on experience with distributed ML frameworks such as PyTorch DDP, Horovod, and Ray Tune, as well as containerized execution with Docker/Singularity and workload orchestration using Kubernetes. This project offers an opportunity to deepen your understanding of scalable AI workflows, HPC environments, and performance optimization in scientific computing.

Speaker: Anjali Khantaal

LT_Anjali_final.pdf

LT_Anjali_final.pptx

14:31 → 14:38 Integrating novel GPU-accelerated Monte-Carlo event generators into the High-Energy Physics toolchain

Monte-Carlo event generators are used in High-Energy Physics (HEP) to simulate collider events. They are a cornerstone in the physics programmes of CERN experiments such as ATLAS and CMS. Given the very high experimental requirements on precision and the strong trend of increasingly relying on hardware accelerators such as GPU for high-performance computing, our group develops a new generation of novel parallelised event generators that are accelerated by GPU devices and CPU vector instructions. This project will focus on the portable parton-level event generator framework Pepper. The student will design, implement, and test C++ and/or Python API for the Pepper framework to provide its deeper integration in the HEP toolchain. This will facilitate the use of accelerated event generation in the production of large-scale simulated event samples for the LHC and enable the use of the framework as a building block for cutting-edge theoretical calculations at next-to-leading and next-to-next-to-leading order in perturbation theory of quantum field theory. In addition, such an interface would facilitate using Pepper for generating training data for modern Machine Learning applications like studying Monte-Carlo sampling based on deep neural networks or training high-fidelity surrogate models for the evaluation of computationally expensive scattering matrix elements.

Speaker: Carla Judith Lopez Zurita

CERNopenlab_CarlaShort.pdf

CERNopenlab_PresentationCarla_pdf.pdf

14:38 → 14:53 Coffee Break

14:53 → 15:00 The ATLAS Anomaly Detection Trigger for Run-3

In 2025 the ATLAS trigger system will be running its first machine learning-based anomaly detection trigger, which includes both L1 and HLT algorithms. The objective of a 2025 summer project will be to study the first data selected by this trigger, and use these results to inform future searches and trigger improvements looking towards 2026 and the HL-LHC.

Speaker: Julia Sophie Troppens (CERN)

The ATLAS Anomaly Detection Trigger for Run 3.pdf

The ATLAS Anomaly Detection Trigger for Run 3.pdf

15:00 → 15:07 Enhancing the Level-0 Muon Trigger System for the High-Luminosity LHC

The ATLAS experiment is upgrading its muon trigger system for operation at the High-Luminosity LHC. The necessary significant improvement in the selectivity of muon tracks within the first-level trigger relies on, for the first time, muon tracking data from precision monitored drift-tube (MDT) chambers.

This research explores the feasibility and benefits of integrating machine learning into the challenging real-time environment of the ATLAS trigger system, aiming to enhance the experiment’s discovery potential in the high-luminosity era. We investigate the use of machine learning algorithms to improve muon reconstruction for the ATLAS first-level trigger. This work involves the development of various neural network models, with algorithms being optimized for potential deployment on powerful FPGA devices. The performance of these models will be evaluated and compared to that of the baseline analytic algorithm in terms of trigger efficiency and muon momentum resolution.

Speaker: Francisco Resende (CERN)

Presentation_Francisco_Resende_openlab.pdf

15:07 → 15:14 Consolidation and automation of CERN's Cloud monitoring dashboards

The CERN Cloud Infrastructure service administrates CERN's private cloud, which provides computation resources to physics analysis and IT services for the whole organisation. At present, this includes around 14.000 virtual machines and more than 10.000 physical ones. The service is built upon OpenStack, a leading open-source software used around the world. As part of our daily operations, it is vital to continuously monitor the health and performance of our systems. This not only gives insight to operators but also provides our users with information about their resources. This monitoring is accomplished using CERN's monitoring platform, with the collected data being displayed by well over 100 Grafana dashboards, such as 1. The goal of this project is to consolidate our many monitoring dashboards into a single software-defined source, which will be easier to update and expand to accommodate future needs. You will write and deploy the Continuous Integration (CI) setup for the creation of such dashboards, and start the development effort of replicating our existing dashboards with Jsonnet/Grafonnet code. This project requires experience with programming, version control systems (git), rudimentary familiarity with Linux, and ideally continuous integration systems (Gitlab CI). Familiarity with Grafana and Jsonnet is a plus, but not required. You will learn about large-scale infrastructure management and gain insight into CERN's cloud platform.

Speaker: Dennis Alexander Mertens Velasquez

CERN_Presentation_final.pdf

15:14 → 15:21 Create a collection of albums in the CERN Document Server from digitized historical photos of particle tracks

Tens of thousands of images of particles tracks from the 1970s have been digitized in 2024 and stored on the CERN Data Cloud (EOS). They are grouped into folders corresponding to the envelopes in which they were stored, and they are named with the information that was printed in the back of each photo. I will have to analyze the set of images, design a process to create albums and records, and finally implement the programs enabling the upload of all the photos into a public collection of the CERN Document Server. It will therefore give these images a new audience, as they will become available online.

Speaker: Danae Broustail

Lightning Talk Danae Broustail(1).pdf

Lightning Talk Danae Broustail(1).pptx

15:21 → 15:28 Profiling of calibrations and evaluation of database solutions for the CMS Alignment and Calibrations system in the context of the NGT-CMS-HLT Optimal Calibrations project and Phase-2 Upgrade for the HL-LHC

Precise knowledge of the CMS sub-detector conditions, alignment and calibrations is crucial for global event reconstruction and significantly impacts physics performance - both at the online High-Level Trigger (HLT) software filtering stage and for the final offline reconstruction. This non-event information is stored and accessed via the CMS offline conditions database, which primarily relies on Oracle. This project aims to evaluate alternative database solutions (e.g. HDF5, CREST, BLOBs, HTTP/CVMFS) in conjunction with CMSSW - particularly in the context of the Next Generation Triggers (NGT) project, which seeks to optimise the calibration procedures for the HLT (Task 3.4), in view of the Phase-2 Upgrade for the HL-LHC.

Speaker: Carson Magnuson Glines

CGLightningTalkFINAL.pdf

15:28 → 15:35 Anomaly Detection within a mission critical ATLAS sub-system

All major CERN experiments along with the accelerator will enter soon in the phase of a big upgrade cycle, called the High-Luminosity LHC, in order to further broaden the physics reach. ATLAS has a plethora of upgrades concerning the HL-LHC era which will equip the detector with many exciting new opportunities. One of the core upgrades in ATLAS concerns the way of reading out the calorimeter sub-detector. In contrast to previous runs ATLAS will be able to read the full granularity of the calorimeter at the level of the hardware trigger system. Having this information available in such a challenging environment provides a unique opportunity to explore Machine Learning ideas on the edge within the context of the ATLAS Global Trigger. The selected student will be able to work on ultra-fast algorithms that use the calorimeter information and be able to reconstruct physics properties of interest against noise that is generated from the secondary collisions present in every bunch crossing (pile-up). The project will explore definitions of more sophisticated physics quantities applied on jet reconstruction, like mass or substructure variables, that can aid specific selection strategies of ATLAS and expand into the area of anomaly detection where Machine Learning has been proved to enormously assist. With this work we aim to further expand the physics reach of ATLAS and ensure that the potential biases introduced during the event selection stages do not affect the quality of physics recorded.

Speaker: Kalle Eemeli Pakarinen (Helsinki Institute of Physics (FI))

Lightning_Talk_Kalle.pdf

15:35 → 15:42 Unified TLS Certificate Automation Using Let’s Encrypt

The Application Hosting service at CERN provide platform for several critical application running at CERN (Single Sign On, Access Control, EDH, IMPACT etc.). A crucial critical component of this platform is the highly available cluster of load balancers that expose these applications to the internet and connect the proxy backends running on multiple Kubernetes clusters. Currently, certificate management for these applications involves manual steps. This project aims to build a unified solution to automate certificate management for both Kubernetes and physical/virtual machines. To achieve this, we will investigate the use of Let's Encrypt to generate certificates. Leveraging Let's Encrypt's ability to issue certificates per host, we will implement Server Name Indication (SNI) in the external reverse proxy. SNI allows the client to specify the hostname it is requesting to the server during the TLS handshake, enabling the server to present the appropriate certificate. This project will enhance the security and scalability of the CERN Application Hosting service by automating certificate management and modernizing the current setup with SNI.

Speaker: Amelie Leconte

CERNLightTalk-AmelieLeconte.pdf

15:42 → 15:49 Development of a CoPilot tool using Large Language Models to assist in editorial and analysis tasks

The CMS collaboration is one of the largest in the field of physics, with its analysis groups being among the most active in data analysis. Each year, the collaboration produces over 100 innovative scientific papers, including the groundbreaking discovery of the Higgs boson in 2012. Physicists from around the globe utilize data collected by the CMS detector to search for new phenomena and test the validity of the Standard Model. In the high energy physics field, employing the most advanced analysis techniques and tools is crucial. This project aims to create a custom Large Language Model (LLM) that can be integrated into the research process to assist CMS physicists with both editorial tasks (such as paper writing, proofreading, and plot drawing) and analysis software design, specifically using the CMS nanoAOD and miniAOD data formats. It will build upon previous work on LLAMA at CERN. Additionally, from a machine learning perspective, optimizing and using an LLM in the demanding scientific environment at CERN presents a fascinating challenge and could be very helpful to the evolution of LLM.

Speaker: Aikaterini Nikou (The University of Edinburgh)

Aikaterini_Nikou_Openlab_talk.pdf