Workshop for potential user partners

• Agenda:
• Quick round-table presentation of participants
• Introduction
• Fundamental motivation behind the centre initiative (Heidi Sandaker, leader of Norwegian Centre for CERN-related Research, and Sveinung Løset, Vice-dean of Research, Faculty of Engineering, NTNU)
• Presentation of centre initiative
• Presentations from potential partners (Statnett, Hydro, Aker BP)
• Open discussions and feedback from participants
• Summary and next steps

Introduction

• Purpose of the workshop:

​

• Better understanding of ideas behind centre initiative for potentially interested user partners

​

• Better understanding of AI needs from a user perspective for research partners

Introduction

• The Research Council of Norway (RCN) will fund 4-6 AI-centres with in total 850 MNOK over 5 years from late 2025
• The application deadline is January 15 2025
• Our initiative is a merger of two outlines submitted for the outline deadline in June 2024:
• one from NorCC (Norwegian Centre for CERN-related Research)
• HAIX : A cross-disciplinary sustainable environment for harnessing AI and exascale data and computing
• one from Faculty of Engineering, NTNU
• KINETIC - Knowledge-Informed and Energy-Efficient Artificial Intelligence for a Disruptive Green Transition

​

​

AI-centre initiative

• Our goal is to unite unique knowledge of applied AI development in one of the world’s most prestigeous research centre communities, CERN, with the strongest environment for technological research in Norway

​

• Achieving by merging fundamental requirements from RCN:
• a centre should be truly national with several research partners
• the core of centre activities should be cross-disciplinary high-quality AI research, applicable to a variety of domains
• research groups both from technology and humanities/social sciences should be involved
• The essence of the two original outlines is still visible in the activity structure of the new centre, and the RCN requirements are met

Research partners

• UiO (Experimental high-energy physics)
• HVL (Dept. of Computer science, Electrical engineering and Mathematical sciences + HVL Business School)
• NTNU (Faculty of Engineering, various departments)
• UiB (Experimental high-energy physics)
• UiA (Dept. of Information and Communication Technology + Dept. of Engineering Sciences)
• CERN
• ETH Zürich (Department of Physics)
• Potential user partners are essential in defining relevant use cases, providing data for algorithm development and testing, etc.

AI-centre initiative

• Funding:

​

• Applying to RCN for the upper limit of 200 MNOK over 5 years
• Significant fraction of the budget will be allocated to 25-30 PhD/postdoc positions

​

• Additional own funding from research partners at least 70 MNOK
• Including 10+ PhD/postdoc positions

Accelerating AI Research

​

For a safe, sustainable and efficient digital transition in Norway

Physics-driven AI Research at the Exascale (HAIEX)

Physics-driven AI research in the Exascale Era�

• Vision:
• Accelerating AI Research

- for a safe, sustainable and efficient digital transition in Norway

• Mission:
• We will contribute to an accelerated safe, sustainable and efficient digital transition in Norwegian businesses and society by leveraging AI knowledge developed in CERN experiments and by the centre partners

Objectives

• Building trustworthy AI systems by better understanding of uncertainties in algorithm predictions
• Using CERN infrastructure, address challenges of large data volume management and complex distributed computing
• Developing fast and energy-efficient AI algorithms for complex decision-making
• Developing physics-informed hybrid algorithms, integrating AI with physics-based approaches
• Understanding risks and vulnerabilities of complex AI systems, predicting and anticipating emerging threats and risks
• Establishing a comprehensive framework for responsible AI development and implementation
• Maximizing impact by collaboration, outreach and education

​

Activities in the centre

• Trust and reliability
• Trustworthiness, robustness, uncertainty quantification
• Fast inference and managing Big Data for AI
• Anomaly detection techniques, use of fast specialized hardware, I/O systems for efficient storage and access, large heterogeneous distributed computing systems
• AI security and privacy
• Understanding potential vulnerabilities of complex AI systems, predicting threats, bias mitigation techniques in AI systems
• Fast and efficient AI for complex decision-making
• Condition monitoring and control, predictive maintenance, energy-efficient AI algorithms for instant real-time predictions
• Physics-informed hybrid modelling
• machine learning algorithms in combination with advanced physics-based approaches, hybrid multi-scale modelling, multi-physics simulation, generative AI for simulation
• Responsibility and legitimacy of AI knowledge
• Risks, dilemmas and societal legitimation/delegitimation in AI-generated knowledge, innovation and the impact of AI technology/knowledge transfer

Preliminary structure of centre activities and impact domains�

Fundamental research using particle collisions at CERN

High-performance event-selection systems (“triggers”)

40 million collision per second

​

60 TB/second

​

24 million 30 Mbps broadband connections

Decision in less than 1/1000th seconds

100 thousand collision per second

​

160 GB/second

​

43 thousand 30 Mbps broadband connections

Decision in 0.5 seconds

thousand collision per second

​

1.5 GB/second

​

400 30 Mbps broadband connections

Simulated data, produced using a digital twin of the experimental equipment

Big Data

> 1 EB on disk/tape

> 1 million cores

2 million tasks/day

Distributed across 170 data centres in 42 countries

​

​

CERN-links to industry, innovation and education activities

NextGen Trigger Project: Eric & Wendy Schmidt Fund for Strategic Innovation to develop AI algorithms to analyze raw data from CERN

Total of $48 million, ~280 FTEs over five years

​

Computing technology developments at CERN has resulted in many technology startups in countries that keep a sufficient volume of researchers involved in CERN activities

​

A long-running tradition of having Master and Bachelor students from NTNU/NorCC in the CERN Technical Student programme

​

PhD students co-financed between NTNU/NorCC and CERN

High-performance event-selection system

Fast inference using AI algorithms on FPGAs

AI in detector and data quality monitoring

HAIEX - Physics-driven Exascale Era AI research

Novel methods of real-time data processing

Big Data@CERN

Generative AI for more efficient simulations

Anomaly detection in large data sets

Uncertainty quantification in AI algorithms

Generative algorithms for industrial applications

Efficient industrial simulation tools for product and process design.

Improve the speed of algorithm execution in order to facilitate complex decision-making in time-critical tasks

Integrating symbolic and sub-symbolic AI to enhance

decision-making in predefined industrial use cases

Norwegian industry and society

AI-relevant technologies and knowledge related to

data management, metadata and resource connection

Education of

MSc and PhD students with CERN and industry

Physics-informed hybrid modelling

Activity: Fast and efficient AI for complex decision-making

Leader: NTNU

​

• Objective: Fast AI-driven decision-making
• Tasks:
• AI-based compression and decompression
• development of graph neural networks for fast decision-making in online monitoring systems
• integrating symbolic and sub-symbolic AI
• Deliverables:
• functional algorithm for compression and decompression
• functional algorithm for decision-making
• decision-support framework established

Improve the speed of algorithm execution in order to facilitate complex decision-making in time-critical tasks

Integrating symbolic and sub-symbolic AI to enhance

decision-making in predefined industrial use cases

Activity: Physics-informed hybrid modelling�Leader: NTNU

Activity: Physics-informed hybrid modelling

Deliverables:

• open source software/plattform
• customized industry-specific solutions (subject to additional financial support)

Activity: AI security and privacy

Leader: UiA

• Objective: create more secure, trustworthy AI models facilitating research, development, and collaboration among stakeholders across ecosystems (digital and business)
• Tasks:
• Task 3a: AI security and privacy algorithmic foundation
• Task 3b: AI security and privacy test and validation platform
• Task 3c: AI security and privacy impact assessment
• Deliverables:
• Sandbox and platform for test and validation of AI models
• AI safety and risk impact assessment framework

Activity: Maximising impact through collaboration �Leader: HVL �

​

Task a: Norwegian collaboration

We will take a leading role in establishing a consortium of all the AI research centres financed through this call

​

Task b: International collaboration

Build sustainable research collaborations on AI across nordic institutions and research groups, initiate and build upon existing relationships.

​

Task c: Education and dissemination

• Plan to supervise 65 Bsc and 75 Msc projects in computer science, software engineering, physics and computational science in course of the center’s initial 5-year lifespan.
• Facilitate student internships at private and public partner companies, at CERN and by partner institutions
• Customized lifelong learning program - courses, training, internships

​

Task d: Centre future�Secure additional funding for the HAIEX center

Activity: Trust and reliability�Leader: HVL

Task a: Uncertainty quantification �Develop mechanisms to quantify and report uncertainties alongside predictions, helping users understand the confidence level of each decision.��Deliverables: Open software, publications�

Task b: Robustness and generalisation�Develop and extend methods for evaluation and improvement of robustness in AI systems. Identify when AI methods are extrapolating outside their knowledge domain and quantify the effect and risks of doing so.

Deliverables: Open software, publications, input to task 1c�

Task c: Guidelines and certification�Develop a national certification system for ethical and safe AI, based on a framework designed in this initiative

Deliverables: Guidelines, certification system, courses and seminars on ethical and safe AI�

Activity: Responsibility and Legitimacy of AI Knowledge�Leader: HVL Business School

Task a: Dynamic Living Lab for Responsible AI Development�Deliverables:

• Stakeholder engagement protocols
• Feedback mechanisms for technical WPs
• Impact assessment methodologies
• Knowledge transfer guidelines from CERN to Norwegian context

Task b: Sectoral Implementation and Knowledge Transfer�Deliverables:

• Assessment criteria for AI implementation across sectors
• Verification methods aligned with Norwegian regulatory requirements
• Integration guidelines for different application domains
• CERN-inspired best practices for each sector

Task c: RRI-AI Integration Framework�Deliverables:

• Combined RRI-AI methodology
• Ethical decision-making support tools
• Best practices guide for responsible AI development
• Framework for measuring and enhancing AI legitimacy