MLDCAT-AP (Machine Learning DCAT-AP) is an application profile that extends DCAT-AP in the field of machine learning. MLDCAT-AP aims to describe machine learning models, together with their datasets, quality measured on the datasets and citing papers.
1. Origin of MLDCAT-AP
MLDCAT-AP has been originally developed in collaboration with OpenML during a pilot. Over time, concepts coming from AI Act and AI Code of Practice (AI Cop) have been included and a reference to them has been added for each class or property where needed, including those references to existing data models.
2. History of MLDCAT-AP releases
- MLDCAT-AP 1.0.0 (13/01/2023) originated from a pilot in collaboration with OpenML and reuses DCAT-AP 2.1.1 concepts.
- MLDCAT-AP 2.0.0 (25/06/2024) takes into account a comparative analysis between repositories of machine learning models, including classes Paper, Algorithm, Data Quality and Risk in view of the AI Office and AI Act.
- MLDCAT-AP 2.1.0 (18/02/2025) was updated to align with AI Act Annexes XI-XIII. This release included, among other elements, Benchmark, Computer Infrastructure, Hardware, Environmental Impact.
- MLDCAT-AP 3.0.0 (30/09/2025) is aligned with DCAT-AP 3.0.1 and adds elements to align with the AI Code of Practice, including the class Modality.
3. MLDCAT-AP pilot with AI4EOSC
AI4EOSC is a research infrastructure project that finished in August 2025. The main aim of AI4EOSC was to develop a platform for building, sharing and deploying AI and ML models for science and research. The platform can also be used with a REST API, that now includes the possibility to get the MLDCAT-AP version of the metadata.
The MLDCAT-AP pilot had the goal to improve the metadata and the metadata management tools in order to publish a machine-readable version of the metadata to improve the overall transparency. The pilot was concluded on 31 August 2025. The mapping to MLDCAT-AP via JSON-LD context is done in an automated way without impacting the users of the AI4EOSC platform.
4. Events
MLDCAT-AP webinar
On 09 September 2025, a MLDCAT-AP webinar was held. More information on the webinar, as well as the published slides, meeting minutes and recording can be found here.
ENDORSE conference
On 08 October 2025, MLDCAT-AP was presented during the ENDORSE conference. The presentation was titles "MLDCAT-AP: An application profile to catalogue machine learning models under the AI Act.
The abstract of the conference presentation can be found below:
"As Artificial Intelligence (AI) and Machine Learning (ML) enter increasingly more and more diverse domains, the need for standardised metadata has risen to ensure interoperability, transparency and compliance. The lack of a standard practice for describing ML models and datasets poses a significant challenge. Without a consistent approach, discovering, understanding, and reusing existing machine learning models and datasets becomes difficult, resulting in duplicated efforts and inconsistencies. Recognising these challenges, SEMIC, in collaboration with OpenML, aimed to develop a specification that would effectively address these issues. The objective was to create an application profile for describing machine learning models, datasets and other relevant concepts. This resulted in the creation of the Machine Learning Data Catalogue Application Profile (MLDCAT-AP).
As part of the Interoperable Europe initiative, MLDCAT-AP addresses challenges in sharing and reusing ML models and datasets across diverse domains and organisations. By leveraging standardised data practices, MLDCAT-AP facilitates the integration of machine learning models into broader data ecosystems, promoting transparency and reproducibility in ML research and development. MLDCAT-AP represents a significant step forward in enhancing the interoperability and discoverability of machine learning models and encourages compliance with the AI Act."
5. Call to action
- Please provide additional feedback on GitHub.
- Contact the SEMIC team to discuss pilot opportunities through DIGIT-SEMIC-TEAM@ec.europa.eu.
5. About SEMIC
Visit the SEMIC collection on the Interoperable Europe portal.