FH SciComp Workshop 2024

1 Jul 2024, 14:00 → 2 Jul 2024, 14:00 Europe/Berlin

Seminar Room 4a/b (DESY Campus Hamburg)

The workshop is organized in the frame of the recently started FH platform on scientific computing (XWIKI FH-SCiComp). Main goals of the workshop are

• Get an overview over the ongoing research activities
• Get to know each other
• Identify synergies

The workshop is primarily intended for PhD students and post-docs, who are actively pursuing the work. Of course also interested staff is invited.

The first part of the workshop is jointly with the Future Colliders workshop, https://indico.desy.de/event/43916/

Topics:

• heterogeneous computing (software exploiting GPUs, FPGAs)
• (generative) Machine Learning & AI applications
• fair data management
• research software engineering (incl. generators and analysis tools)
• quantum computing
• DAQ , online & real time applications, trigger
• Collaborative tools
• Other cool topics we have missed
•  

Please register to help us organizing the breaks and the reception on Monday after the session.

Please submit a short abstract to help us organizing the program.

More details will follow soon. Ideally subscribe to the mailing list of platform fh-scicomp@desy.de (via the DESY mailing list server).

 

 

Monday 1 July

Scientific Computing for Future Colliders

1 Event Generators for Future Colliders

Speaker: Juergen Reuter (DESY)

2024_DESY_FC_SciComp_MC_JRR.pdf

2 Discussion

3 Key4hep - Core, Sim and Rec Tools for Future Colliders

Speaker: Thomas Madlener (FTX (FTX Fachgruppe SFT))

key4hep_fh_sciomp_fc_day_01072024.pdf

4 Discussion

5 Applying Machine Learning to Future Colliders

Speaker: Frank Gaede (FTX (FTX Fachgruppe SFT))

gaede_ml_fc_key4hep.pdf

6 Discussion

15:30 Coffee break

Scientific Computing I

7 The SciComp Platform - Mandate, Mission, and Activities

Speaker: Christoph Wissing (DESY)

scicomp_workshop_intro.pdf

8 DESY and the NFDI

DESY plays a significant role in the German national research data infrastructure, with the PUNCH4NFDI and DAPHNE4NFDI consortia both being led by DESY scientists, and with strong contributions to their science goals. The NFDI consortia are now half-way through their first funding period of five years, and deliberations and discussions are commencing on the future setup of PUNCH4NFDI and DAPHNE4NFDI in a potential second funding period.
In this presentation, we give a short introduction to the NFDI and to PUNCH4NFDI as the consortium of particle, astroparticle, hadron&nuclear, and astrophysics, and we discuss potential future directions and projects. We also invite for discussion on and contributions to the shaping of a future programme for PUNCH4NFDI.

Speaker: Christiane Schneide (FTX (FTX Fachgruppe SFT))

20240701_PUNCH_FHSciComp.pptx

9 Belle II detector alignment with Millipede II

The Belle II experiment at the SuperKEKB accelerator aims at precision measurements in B, tau and charm physic sectors. Many such measurements rely on high precision vertexing and thus precise alignment of the detection elements. A global alignment method utilizing Millepede II software package will be presented, which determines around sixty thousand alignment parameters simultaneously. Computational requirements as well as complex time-dependence of the alignment of the newly installed pixel sub-detector pose new challenges for future operation, which will be briefly discussed.

Speaker: Tadeas Bilka (BELLE (BELLE Gruppe))

Belle2AlignmentSciCompDESY1July2024.pdf

10 Solving the CMS High Granularity Calorimeter's Data Processing Challenges with Heterogeneous Computing

CMS is upgrading its endcap detectors for the High-Luminosity LHC runs with a first-ever silicon/scintillator tracking calorimeter: HGCAL. It will allow to handle increased collision pile-up, improve the precision of measurements, and also enable completely new ways to search for new physics thanks to its particle-pointing and timing capabilities. However, these wonderful improvements come at a price - with over 6 million readout channels, processing the HGCAL's data becomes a significant challenge from the computing point of view. In this talk, I will introduce this future calorimeter of CMS, explain related computing challenges, as well as present how we are addressing them using heterogeneous computing.

Speaker: Jeremi Niedziela (DESY)

hgcal_fh_computing_1_07_2024.pdf

11 Generative Machine Learning at CMS

The CMS Generative Machine Learning Group will showcase three distinct projects, each utilizing point cloud-based generative models to advance particle physics research. The first project, "Attention to Mean Fields for Particle Cloud Generation", features an attention-based generative model that adeptly processes complex collider data represented as point clouds, demonstrating effectiveness on the JetNet150 and CaloChallenge datasets. The second project, "DeepTreeGAN", explores novel techniques for iterative up- and downscaling of point clouds, inspired by the tree-based development of particle showers. Finally, "CaloPointFlow" presents a generative model using normalizing flows.

Speaker: Simon Patrik Schnake (CMS (CMS Fachgruppe Searches))

CMS_Generative_Group.pdf

12 ASAPO: A high-speed streaming framework to support an automated data-processing pipeline.

Modern scientific experiments often generate large amounts of data, posing challenges for real-time processing and analysis. ASAPO, a high-performance streaming framework developed at DESY, addresses these challenges by providing a robust solution for online and offline data processing. Leveraging TCP/IP and RDMA over Ethernet and Infiniband, ASAPO facilitates high-bandwidth communication between detectors, storage systems, and analysis processes.

ASAPO offers user-friendly interfaces for C/C++ and Python on all major platforms, streamlining the development of data processing pipelines. A high-level Python library reduces boilerplate code and enables the creation of complex analysis workflows with ease. Key features include automatic retransfer, trivial parallelization on a per-image basis, support for multi-module detectors, and web-based monitoring capabilities.

Several experimental facilities at Petra III already benefit from ASAPO, employing it in various data-processing pipelines. Examples include azimuthal integration of X-ray scattering data, peak finding, and indexing of diffraction patterns. These applications demonstrate ASAPO's versatility and effectiveness in accelerating scientific discovery.

Speaker: Mikhail Karnevskiy (IT (IT Scientific Computing))

ASAPO.pdf

13 From testing cosmological inflation models to solving PDEs -- dNNsolve: an efficient NN-based PDE solver

!!!Please place me into the Monday afternoon 4:30 pm - 6 pm session!!!

Neural Networks (NNs) can be used to solve Ordinary and Partial Differential Equations (ODEs and PDEs) by redefining the question as an optimization problem. The objective function to be optimized is the sum of the squares of the PDE to be solved and of the initial/boundary conditions. A feed forward NN is trained to minimise this loss function evaluated on a set of collocation points sampled from the domain where the problem is defined. A compact and smooth solution, that only depends on the weights of the trained NN, is then obtained. This approach is often referred to as PINN, from Physics Informed Neural Network. Despite the success of the PINN approach in solving various classes of PDEs, an implementation of this idea that is capable of solving a large class of ODEs and PDEs with good accuracy and without the need to finely tune the hyperparameters of the network, is not available yet. In this paper, we introduce a new implementation of this concept - called dNNsolve - that makes use of dual Neural Networks with different activation functions to solve ODEs/PDEs. We show that dNNsolve is capable of solving a broad range of ODEs/PDEs in 1, 2 and 3 spacetime dimensions, without the need of hyperparameter fine-tuning.

Speaker: Alexander Westphal (T (Cosmology))

FHH-SciComp_Westphal_2024.pdf

14 Infrastructures for scientific computing & sutainability at DESY-IT

Within the EU-Project RF2.0, DESY focus on sustainable scientific computing infrastructure splitted into more classical/obvious directions like, new architectures, longer system lifetimes, software efficiencies etc. and the model to operate a decent amount of compute resources with variable power consumption coupled to 'true RE power' availability at DESY location by steering the compute load - with dependencies amongst them. Beside these core technical challenges, which we believe are manageable within the project team, the true effective GHG emission reduction strongly depends on user acceptance. In other words - assume that will all be done and works, you will see job completion times extended (vary) and shifted job starting times (even if compute resources are available) due to unavailable RE power:
- what will be the motivation for each user to accept that ?
- i.e. bonus, directorate directive, being a good citizen, ...

Speakers: Christian Voss (DESY), Martin Gasthuber (IT (IT Scientific Computing)), Yves Kemp (IT (IT Systems))

IDAF-FH-SciCompWS.pdf

15 Quantum algorithms for charged particle track reconstruction in the LUXE experiment

We investigate the potential of quantum computers for pattern recognition in track reconstruction at LUXE, based on a quadratic unconstrained binary optimisation and a quantum graph neural network.

Speaker: Yee Chinn Yap (FTX (FTX Fachgruppe SLB))

20240701-YeeY-SciComp.pdf

18:30 Evening reception and get-together

Tuesday 2 July

Scientific Computing II

16 Precise Quantum Angle Generator Designed for Noisy Quantum Devices

The Quantum Angle Generator (QAG) is a new quantum machine learning model designed to produce precise images on current Noise Intermediate Scale (NISQ) Quantum devices. The QAG model uses variational quantum circuits as its core, and multiple circuit architectures are evaluated.
This study explores the QAG model's noise robustness through an extensive quantum noise study. The results indicate that the model, when trained on a quantum device, can learn the hardware noise behavior and produce excellent outcomes. When simulated quantum hardware noise is included, the model's results remain stable until approximately 1.5% of noise during inference and almost 3% in training. However, running the noise-less trained model on real quantum hardware leads to a decrease in accuracy. If the model is trained on hardware, it can learn the underlying noise behavior, where the same precision is achieved by the noisy simulator. Additionally, the training showed that the model can recover precision even with significant hardware calibration changes during training with an increase of noise up to 8% for one qubit.
This work demonstrates the QAG model's ability to learn hardware noise behavior and deliver accurate results in the presence of realistic noise levels expected in real-world quantum hardware. The QAG model is utilized on simulated calorimeter shower images, which are employed in high-energy physics simulations to determine particle energies and to identify unknown particles at CERN's Large Hadron Collider.

Speaker: Saverio Monaco (CMS (CMS Fachgruppe Searches))

FHSciComp-Monaco.pdf

17 Quantum computing @ Scientific Computing

An overview of my ongoing research projects with CQTA Zeuthen and a brief overview of where QC could be used in SC in the future.

Speaker: Lukas Mansour (DESY IT Dual)

Lukas_Mansour_QC_at_SC_v3.pdf

18 Summary of Scientific Research and Innovation Group (IT-RIC) activities

I would like to report on activities of the IT-RIC group in supporting DESY scientific domains. This includes but is not limited to our work on:

• using Artificial Intelligence for advancing imaging in scientific applications;
• providing an open science platform including a catalogue for open data at DESY;
• providing a Helmholtz-wide Scientific IT Infrastructure, in collaboration with other Helmholtz Centers, for sharing compute and storage resources;
• providing Data Transfer and Data Orchestration Solutions to connect to Sciences in Europe based on WLCG technologies including access to European HPC centres.

Speaker: Patrick Fuhrmann (IT (Research and Innovation in Scientific Co))

2024-07-01-FHSciCom-Workshop-Intro.pptx

HI_for_FH_scientific_comp.pptx

19 4D track reconstruction with quantum algorithms

We investigate the use of quantum algorithms to reconstruct charged particle trajectories from 4D tracker measurements.

Speaker: David Spataro (FTX (FTX Fachgruppe SLB))

FH SciCom.pdf

20 Tracker alignment of the CMS detector with Run 3 data

The tracking system of the CMS experiment is the world’s largest silicon tracker with its 1856 and 15148 silicon pixel and strip modules, respectively. To accurately reconstruct trajectories of charged particles the position, rotation and curvature of each module must be corrected such that the alignment resolution is smaller than, or comparable to, the hit resolution. This procedure is known as tracker alignment and will be described in the context of this talk with a focus on performance during the start of Run 3 data-taking.

Speaker: Dr Henriette Petersen (CMS (CMS Fachgruppe TOP))

FH_computing_workshop.pdf

10:30 Coffee break

Scientific Computing III

21 CNNs and GNNs for tagging anomalous showers with ATLAS

This talk will discuss recent work towards developing a ML (CNN/GNN) tagger to distinguish anomalous showers caused by the decays of long-lived particles from QCD jets with the ATLAS detector

Speaker: Lukas Bauckhage (ATLAS (ATLAS Upgrade))

FHSciComp_Bauckhage.pdf

22 KAI-Project: Self Adaptive dCache

dCache is a storage system for scientific
institutions that manages large amounts
of data in a distributed system. Although
the dCache code base already has self-
regulating mechanisms, there are still
situations in daily operation, in which
the storage system is overloaded. The
challenge of the project is to
identify indicators, which can be used to
reliably identify the times and causes of
overload. The focus here is on the
identification and manipulation of
relevant system parameters.

Speaker: Sandro Grizzo (IT (IT Scientific Computing))

Self_Adaptive_dCache.pdf

23 Monte Carlo algorithms for lattice QCD simulations

In this talk, I will discuss the basic ideas of Monte Carlo algorithms for computing correlation functions and how to extract physical information from them. In particular, I will discuss our recent use of advanced algorithms to reduce the statistical errors.

Speaker: Lorenzo Barca (Z_ZPPT (Zeuthen Particle PhysicsTheory))

fhscicomp24_barca.pdf

24 Open Data and Infrastructure Portal for DESY and HIFIS

Following FAIR principles, we are setting up portals for access to Open Data and the corresponding infrastructure to explore the published data sets. The open datasets and their metadata are available in a Scicat instance to everyone interested and can either be downloaded directly or accessed through an open infrastructure portal called VISA upon reasonable request.
In the short talk I will present and architecture overview and short interaction examples with regard to the portals. A blueprint architecture of this setup will be available later this year in a comprehensible form for the scientific community.

Speaker: Tim Wetzel (IT (Research and Innovation in Scientific Co))

2024-07-02-SHSciCompWS-OpenDataAtDESY-V2.pptx

25 Automation of the flavour tagging calibration software in the ATLAS experiment

Particle cascades originating from quarks and gluons decays (jets) are omnipresent in proton-proton collisions at the LHC. The identification of jet flavours is essential for many physics searches at the ATLAS experiment. This is achieved using machine learning algorithms (taggers) which combine tracks and jets information to determine the flavour of the jets ($b$-jets, $c$-jets and light jets). These taggers are trained with simulated Monte Carlo events and, due to simulations imperfections, their performance need to be measured in data in order to extract correction factors for the simulation predictions. ATLAS developed a set of calibration techniques for different jets flavours to correct, then the correction factors need to be re-derived every time a new tagger is deployed. Automating the calibration workflow significantly accelerates the calibration cycle and makes it less prone to manual mistakes. We present the first automated calibration framework in ATLAS using REANA platform. The results are compared with the official results using 36.2 $\mathrm{fb}^{-1}$ of 13 TeV collisions data from ATLAS, and a new set of calibration results with a customised setup is also included. The same method can be applied in other contexts to reduce the amount of time and resources needed to achieve the scientific goals.

Speaker: Marawan Barakat (Z_ATLAS (Experiment ATLAS))

FHSciComp_FTAG_Marawan.pdf

26 MUSICS

The main objectives of the project are to to develop a user friendly service via which users will be able to publish open data following the FAIR principles and also to find these published data again and access them in a convenient and performant way. The service shall be operable via
a large language model based user interface so that users would be able to ''talk'' to the service requesting the required actions. Behind the scenes the requests would be translated into search commands to the metadata catalogue and the results can be confirmed by the user. After the
URL to the storage infrastructure has been given an automatic redirect to the corresponding storage infrastructure would be provided and the data are being served. This can be accomplished by combining the already existing components at DESY with the locally developed distributed storage middleware ''dCache'' being in the centre of it. Further major components are the SciCat based metadata catalogue, the Helmholtz Authentication and Autorisation infrastructure as well as an easy to use landing page as user interface and for binding the components together.

Speaker: Dr Kilian Schwarz (IT (IT Scientific Computing))

MUSICS.pdf

27 CaloClouds: Fast Geometry-Independent Highly-Granular Calorimeter Simulation with Diffusion Point Clouds

A novel approach for fast simulation of energy depositions in high-granular detectors, essential for upcoming high-luminosity collider experiments.

Speaker: Anatolii Korol (FTX (FTX Fachgruppe SFT))

CaloClouds_SciComp_Anatolii_Korol.pdf

Discussion