The key expected outcomes include:

• Realistic applicability of AI at large scale in 6G networks for natively supporting AI architectures, common data sets and/or federated learning methodologies and assessment models, including re-training of models with the introduction/update of the data sets; AI/ML solutions that will have impactful contribution to standardisation activities; Interpretability solution exploring standard-compliance testing & debugging techniques.
• Development of curated data sets of realistic 6G scenarios (using new real and/or synthetic data sets) for reference usage in telecommunication research and standardisation, targeting their wide acceptance and future usage for benchmarking by future EU R&I activities.
• Analysis, aggregation and harmonisation of results from existing projects and creation of an overall framework for benchmarking and calibration, end-to-end testing and evaluation of AI solutions for 6G networks.
• Metrics and models to assess the pros and cons of AI technologies in telecommunications, including aspects as energy efficiency, explainability, reliability, safety and security, non-discrimination, privacy and performance as well as usability & accessibility for users. Specific focus should be on energy-efficiency and computational complexity that are still open issues for real-time hardware.
• Recommendations for policy and regulatory guidelines on the development and usage of AI solutions for network optimisations and provision of AI as a service.
• Development of a trustworthy AI framework which should be addressed in each stage of the AI system building (from data to model development etc.).
• Focus should be on implementation and connected to current standardization efforts and state-of-the-art Open Source frameworks and tooling.

Please refer to the "Specific Challenges and Objectives" section for Stream B in the Work Programme, available under ‘Topic Conditions and Documents - Additional Documents’.

The focus of this Strand is on several complementary issues and applicants may select several or all the below-mentioned issues. The main goal of this project is to fill the gaps and work on the end-to-end system integration of SNS AI/ML solutions, or national level developed AI/ML solutions and not to focus on dedicated AI/ML problems of specific network domains. The targeted project scope includes:

• Development of a reference framework for end-to-end AI usage for the telecommunications domain in relation to 6G, including methodologies for centralized, distributed and federated applications, reference use cases, data acquisition and generation, repositories, curated training and evaluation data, as well as the technologies and functionalities needed to use it as a benchmarking platform for future AI/ML solutions for 6G networks. The framework should be expandable so that future R&I actions can follow its directives and easily provide new use cases and data sets. Towards this end, the reference framework shall be hardware-agnostic, so that it can support heterogeneous hardware implementations.
• Development of appropriate data infrastructure and functionalities that will enable novel AI-based services as well as AI as a Service to vertical industries.
• Models for AI costs and benefits in telecommunications applications. Typical 6G metrics should be able to be evaluated, including but not limited to data rate, latency, density, energy efficiency, flexibility and performance, and/or security and privacy, but other value metrics can be considered as well.
• Solutions that will guarantee reliable use of the technology and build trust in 6G and services enabled by 6G. Associated topics include: i) AI environment (training, development, production) evaluation; ii) assessment models of reliable AI costs and performance value; iii) conflict resolution among local and global AI models, iv) Vulnerability assessment of AI models for different telecommunication applications potentially using friendly hacking means and v) Reliable and trustable AI life cycle, including the AI development and deployment environments.
• The framework should address a wide range of open issues indicatively and not limited to, e2e AI/ML conflict resolution, placement of AI at appropriate places inside the network (e.g., edge), provide energy friendly AI/ML solutions, how to handle vast amount of data for AI/ML purposes using computing/storage and network resources in a scalable way, and any other advances needed to support the overall goal. In addition, the AI/ML should be able to work across different/multiple network infrastructures, tools, apps, and data/communication needs.
• Where relevant, harmonisation/coordination with Member States or Associated countries 6G initiatives, as well as with the existing SNS EU-US cooperation initiative (HORIZON-JU-SNS-2023-STREAM-B-01-06: EU-US 6G R&I Cooperation). Any produced PoCs should be implemented in a way that their integration in future SNS WP2025-26 Stream C and/or Stream D project will be possible (e.g., open-source solutions, appropriate documentation, support after the completion of the project etc.).
• Production of data sets should cover as many areas as possible from the actual operation of 6G networks (user mobility patterns, RAN/Transport/Core data traffic patterns, network failures or security attacks, computing usage patterns etc.) including real and synthetic data, or even appropriately adapted data from open free data sets.
• Production of data sets and validation methodologies, contributing to 6G Human Centricity and Societal acceptance and in compliance with the rules of data legislation. Development of guidelines, for ethical considerations, and suggestions to regulatory frameworks are also desirable. Methods of accreditation of usage/compliance may also be considered to validate techniques of dataset production and dataset conformance.
• Development of solutions that will address the need for robust and trustworthy AI/ML validating the “quality” datasets from different scenarios, which influences the outcomes of the AI systems, as well as the corresponding outcome of AI.
• Verification and validation of AI techniques over experimental platforms, additionally providing the associated datasets.

Applicants are expected to provide details on the type and availability of the datasets to be produced and curated by the project. This includes, but is not limited to, whether they will be based on existing or new datasets, project partner(s) in charge of producing them, whether they will be based on real-world measurements or synthetic ones, etc; as well as their complementarity, availability of datasets beyond consortium partners.