Expected Outcome:
- Improved methods and tools to optimally develop and deliver bias-free AI algorithms and generative AI.
- New concepts of the auto-evolution-enabled software elements that AI algorithms can autonomously select and integrate.
Scope:
The latest developments in AI are demanding computation infrastructures designed to maximize the number of FLOPS. Europe has a window of opportunity to leverage open source and ensure that the European industry is at the cutting edge of these new processing infrastructures. The current methods and tools to develop and efficiently deliver AI pipelines and complex generative AI applications present several shortcomings.
Current identified challenges include the continuous management of data pipelines, novel testing methods (e.g. differential testing or improved performance testing), optimized deployment strategies (in terms of using energy efficient resources or the best performant), management of dependencies with a diverse set of types of hardware, algorithm bias and discrimination against certain groups, determined by characteristics such as gender, ethnicity, age or disability, as well as the maintenance of the effectiveness of AI applications, notably generative AI ones. The explainability of AI models is another crucial challenge that must be also tackled at the software engineering level, possibly in synergy with explainability methods at the machine learning stage. Also, large language models and foundational models require the development and operation of complex system architectures that need to handle data processing at large scale, continuous training of models and inference. Furthermore, AI pipelines and generative AI application also strongly suffer from poor protection against learning dataset poisoning, as well as prompt poisoning; software engineering methods and tools providing support to protect against those attacks are thus direly required. This presents novel challenges for developers that will need to be addressed with the development of new methods, mechanisms and tools covering the above, including neuromorphic computing, but not limited to.
The main objectives for the advancements of Software Engineering in this field are:
- Increasing the productivity of generative AI application developers, and operators, especially of foundational and language models.
- Define a reference architecture and framework for generative AI application developers to simplify integration and system modularity.
- Simplifying and automating the development and operation of such applications.
- Including novel techniques for the generation of labelled training sets for reinforcement learning (RL) from human feedback, to be evaluated with corresponding RL algorithms.
- Facilitating AI progress and advancement beyond the state-of-the-art in the EU at a faster rate vis-à-vis the rest of the world and contributing to the implementation of the AI Act.
- Establishing collaboration with EU-based chips designers to maximize how the hardware resources are fully optimized with the software to develop
The proposal should address at least one use case as an industry application (e.g., automotive, health, energy, food/agriculture, etc.).
Actions could build on, provide support or seek collaboration with existing projects, develop synergies and ensure complementarities with other relevant European, national or regional initiatives currently demanding this kind of optimisations possibly in different verticals: e.g. digital twins leveraging GenAI, data quality enhancement leveraging GenAI, development of energy-efficient AI algorithms.
Communicable results should be shared with the European R&D community through the AI-on-demand platform, and if necessary, other relevant digital resource platforms to bolster the European AI, Data, and Robotics ecosystem by disseminating results and best practices.
This topic implements the co-programmed European Partnership on AI, data and robotics (ADRA), and all proposals are expected to allocate tasks for cohesion activities with ADRA, and the CSA HORIZON-CL4-2025-03-HUMAN-18: GenAI4EU central Hub.
Actions are encouraged to build on, provide support or seek collaboration with existing projects, develop synergies and ensure complementarities with other relevant European, national or regional initiatives currently demanding this kind of optimisations possibly in different verticals: e.g. digital twins leveraging GenAI, data quality enhancement leveraging GenAI, development of energy-efficient AI algorithms. In particular, proposals are encouraged to ensure complementarities with projects funded under the following topics:
- HORIZON-CL4-2024-HUMAN-03-01: Advancing Large AI Models: Integration of New Data Modalities and Expansion of Capabilities
- HORIZON-CL4-2024-HUMAN-03-02: Explainable and Robust AI (AI Data and Robotics Partnership) (RIA)
- HORIZON-CL4-2023-DATA-01-01 AI-driven data operations and compliance technologies (IA)
- HORIZON-CL4-2021-DATA-01-01 Technologies and solutions for compliance, privacy preservation, green and responsible data operations (RIA)
- HORIZON-CL4-2021-HUMAN-01-24 - tackling gender, race and other biases in AI.
Projects are encouraged to develop synergies and relate to activities and outcomes of the Digital Europe Programme (DEP) and any existing or emerging Important Projects of Common European Interest (IPCEI) initiative.
