Interoperability is Key: How AIoD facilitates connectivity and ease of information exchange

The AI on Demand (AIoD) platform aspires to provide a comprehensive ecosystem designed to provide AI resources, services, and collaboration opportunities to its end-users. Towards the realization of this vision, the platform’s Metadata Catalogue plays a crucial role, as it forms the information and knowledge backbone of the platform.

To reflect the associations between actors, stakeholders, and offerings of the European AI Ecosystem, the Metadata Catalogue must define and employ a comprehensive structure for defining these linkages and their meaning. What’s more, the platform must ensure interoperability, that is, facilitate information exchange with other environments and allow for the easy ingestion of relevant metadata from other platforms connected with AI tools and activities.

For this purpose, AIoD’s Metadata Catalogue relies on the AIoD Conceptual Model, a collection of semantic resources that specify the entities participating in the European AI ecosystem, the relations between them, and the properties that adequately describe all these entities. To model these, the model adopts the core notion of an AI Resource, which is further analysed in AI Assets (i.e. different types of technical and informational assets accessible and usable from the interested audience) and Ecosystem Resources (i.e. entities that are not assets per se but are closely related to them and can be sources of information or guidance, such as people and institutions, projects, and networks). Assets defined in the model include datasets, AI models, experiments, documentary, and multimedia resources, while Ecosystem Resources entail event information, project information, news items including job offerings and funding opportunities, as well as organizations and individuals active in the broader AI ecosystem.

For each of the aforementioned entities, an extensive set of properties is defined in order to provide all the information that can be useful for assessing the relevance of a resource for different use cases. When applicable, such information is further structured via the usage of the appropriate taxonomies for assigning values to the relevant properties, e.g. for specifying the core business sector where an organisation is active, the business function and specific business problem addressed by an AI model, and the research area to which a solution pertains.

From a technical standpoint, it is important to note the adoption of semantic web standards for expressing and deploying the model. These standards allow the effective introduction of the model in the platform’s development process, as there are multiple readily available tools and services for parsing and exploiting the model. Furthermore, semantic web standards enable the straightforward linkage of the model’s entities and terminology with important external standards used by data spaces, specialized sector-targeted communities, and important platforms like Google or Hugging Face. Moreover, any existing or future external platform and repository can be integrated with the AIoD Platform via connector applications that carry out the mapping between the AIoD model and the representation schema adopted by the external platform.

In summary, the AIoD Metadata Model and its application to the AIoD Metadata Catalogue ensures that the collected information from participants in the European AI Ecosystem is easily ingested, maintained, updated, and interlinked. Thus, it constitutes a critical factor for the evolution, sustainability, and continuous relevance and prominence of the platform.

For more information and for following the evolution of the model, please visit the dedicated GitHub repository that hosts the model and its documentation.