Algorithms for AI – High-performance computing (WP1)
WP1 establishes the methodological, algorithmic and data foundations for the development of reliable, scalable and trustworthy artificial intelligence across the entire NCUI project. It focuses on advanced AI models, high-performance computing (HPC), working with large-scale data, and new mathematical approaches to understanding neural networks and transformers.
At the same time, it is developing infrastructure for the effective training and deployment of AI models in sectors such as industry, energy, autonomous systems, construction, digital libraries and medicine. An integral part of this is also research into the transparency, robustness, security and legal aspects of AI, which are key to its trustworthy use in practice.
What problem does the WP solve, and why is it important?
01
Today’s AI models are becoming increasingly complex, demanding more data and computing power, and difficult to explain
02
There is a lack of a unified environment for working effectively with the large volumes of diverse data required for training AI
03
Many AI systems are not sufficiently robust against errors, changes in the environment or deliberate manipulation
04
Organisations need transparent and reproducible processes for the development, testing and deployment of AI
05
New European AI regulations are increasing the need for methods to assess reliability, safety and compliance with legal requirements
06
Advanced applications, such as autonomous driving, energy forecasting and digital twins, require the integration of AI with physical models and high-performance computing
Examples of use / areas of application and benefits
The automotive industry and autonomous mobility
The creation of realistic 3D scenes and virtual worlds for testing driver assistance and autonomous systems.
Energy secor
More accurate forecasts of energy generation and consumption, detection of anomalies in energy networks, and support for risk management
Industry and Manufacturing
More efficient training and operation of AI models to optimise manufacturing processes
Digital Libraries
Automated processing, classification and intelligent search in large-scale text, image and multimedia archives
Robotics and Navigation Systems
Improved 3D modelling of the surrounding environment using image and LiDAR data
Public administration and AI regulation
Methodologies and tools for the safe, transparent and legally responsible use of artificial intelligence
Key people
Participating institutions
Used technologies and procedures
- Large language models, transformers and deep neural networks
- Neuroalgebraic geometry for the analysis and interpretation of AI models
- Physically informed neural networks (PINNs)
- Graph neural networks (GNNs) and neural operators
- Geo Foundation Models
- 3D reconstruction, computer vision and Neural Radiance Fields (NeRF) technologies
- High-performance computing (HPC), cloud platforms and distributed data processing
- Workflow orchestration and MLOps (Apache Airflow, HyperQueue, MLflow, DVC)
- Data provenance, auditability and reproducibility of AI processes
- AI evaluation methods based on game theory, security and system trustworthiness
- Analysis of the ethical, legal and societal impacts of AI
- High-performance computing (HPC), cloud platforms and distributed data processing
- Workflow orchestration and MLOps (Apache Airflow, HyperQueue, MLflow, DVC)
- Data provenance, auditability and reproducibility of AI processes
- AI evaluation methods based on game theory, security and system trustworthiness
- Analysis of the ethical, legal and societal impacts of AI
Achieved and planned results
2027
Experimental software for 3D reconstruction (version 1)
R- software
Tools for measuring strategic behaviour, situational awareness and an AI’s ability to recognise the test context
R- software
Participatory platform for VSD
R- software
A prediction system for the energy sector based on advanced neural networks
Gprot – prototyp
2028
A tool for processing library data, optimised for use in AI processes
R- software
A game-theoretic library for modelling AI evaluation, focused on AI systems capable of strategic reasoning
R- software
2030
Experimental software for 3D reconstruction (version 2)
R- software
2031
A framework for training AI models focused on digital libraries using HPC infrastructure
R- software