deliverable report

3.4 Automatic generation of AI solutions that respect EU regulatory framework (M5-M33)

This deliverable will present the initial AI tools that will be produced from automatic generation of AI models and multilingual interaction with the end-users.
Task T3.4 is strategically positioned to revolutionise the landscape of AI algorithmdevelopment by introducing automated processes that ensure unbiased models while adhering to the EU regulatoryframework. Leveraging state-of-the-art techniques such as AutoML, transfer learning, and meta-learning, this taskfocuses on the efficient selection of AI models and pipelines. By automating the model recommendation process, theplatform ensures optimal model performance for identified use cases, promoting efficiency and accuracy in algorithmselection. To democratise the use of AI algorithms, the task involves the automatic generation of wrapper code foridentified use cases. This includes facilitating zero-shot use of Large Language Models (LLMs) and adapting or fine-tuning algorithms on new data. The goal is to make AI development more accessible and user-friendly, promoting theadoption of advanced AI technologies across diverse domains. Furthermore, recognizing the paramount importance ofaligning with regulatory frameworks, the task ensures that the AutoML processes developed within ALFIE complywith existing and upcoming EU regulations, particularly those outlined in the AI Act. This includes robust measuresto address ethical considerations, data privacy, and transparency in algorithmic decision-making. Beyond compliance,T3.4 takes a proactive stance by proposing policy measures that support the responsible and ethical use of no-code AIand related tools. This involves providing guidelines and recommendations to policymakers to foster an ecosystemwhere AI technologies contribute positively to society while mitigating potential risks. Roles: TU/e will lead the task deploying the appropriate tools for the auto-generation of AI tools, EHU will support data privacy and connectionwith the Semantic Knowledge Graph. KInIT will supervise the task as WP leader and provide guidelines to connectwith the interaction layer. UoB will be responsible to connect the auto-generation layer with evolving AutoML toolsT3.5 Evolving AutoML algorithms by involving humans in the loop (M5- M33) [lead: UoB; contrib: DAS, TU/e]

4.2 User-centric platform design of system architecture and orchestration layer

This deliverable will contain the technical requirements, considered during the implementation of the AutoML platform. It will also describe the functionalities that will be supported by the platform, including the security framework. Additionally, the deliverable will present the UI and UX prototypes that will be deployed for AutoML purposes. Finally, it will outline the platform architecture that will be used for the implementation of the platform prototypes

Requirements

Initial AutoML platform co-designed architecture, functional requirements and operational prototype|This deliverable will contain the technical requirements, considered during the implementation of the AutoML platform. It will also describe the functionalities that will be supported by the platform, including the security framework. Additionally, the deliverable will present the UI and UX prototypes that will be deployed for AutoML purposes. Finally, it will outline the platform architecture that will be used for the implementation of the platform prototypes

This task will design and describe the overall architecture of the AutoML platform,building upon (i) the selection of the best practices, algorithms and approaches that will be identified (T2.5, T2.6), (ii)the user’s requirements and needs (T2.3, T4.1), and (iii) the defined scenarios and use cases (T4.5), following also theUCD’s principles (T4.1). The task will start by performing a mapping of the user requirements to system functionalitiesand technical specifications, ensuring that the latter is in line with the prioritisation of T2.5, T2.6. The outcome of thisactivity will be the overall system architecture, its break-down into functional and non-functional modules andcomponents, with detailed specifications for each of them, including also and the operational requirements, as well asdetails over their interactions and software interfaces, to allow their independent development. Equally importantly,the architecture shall address diversity in equipment (hardware and software) and resources, to allow the envisagedplatform to operate in diverse H/W setups, while also enabling future changes, updates, and upgrades while at thesame time supporting the design operations continuum of dependable AutoML platform. T4.2 also aims to lay thefoundation for the AutoML platform by implementing an orchestration layer that will seamlessly integrate componentsfrom WP3 and WP4 (e.g. interfaces). The emphasis will be placed on creating a lightweight yet stable platform thatwill incorporate essential layers such as: i) the understanding and interaction layer(T3.2) that efficiently processes userinput and translates it into a structured format and acting as the interface through which users interact with the AutoMLplatform, ii) the Automated machine learning (AutoML) (T3.4) that allows the user to autonomously select suitablemachine learning models and generate code for their deployment; iii) the trustworthiness layer (T3.6) that isresponsible to integrate the platform with the appropriate XAI tools to ensure transparency and user understanding.Roles: TU/e will lead the task, designing the architecture of AutoML platform and the implementation of theorchestration layer. CERTH and KInIT will help TU/e to identify and integrate all technical layers in the AutoML.