Machine Learning Lifecycle Ontology Working Group
Get Involved
Overview and Purpose
The Machine Learning Lifecycle Ontology (MLLO) Working Group is developing a comprehensive, modular ontology designed to establish a shared semantic foundation for consistent and description of machine learning (ML) artifacts, processes, datasets, algorithms, and quality metrics throughout the entire ML lifecycle from data preprocessing to model deployment and retrain. This initiative is driven by the rapid adoption of ML across industries, the increasing complexity of ML systems, the growing need for reproducibility and transparency of ML models, and the challenge of integrating ML workflows across diverse tools, platforms, and organizational contexts.
Current ML lifecycle management faces significant challenges in documentation, reproducibility, traceability, and knowledge sharing. ML artifacts, including datasets, models, training configurations, evaluation metrics, and deployment specifications, are created, maintained, and utilized across heterogeneous systems such as data preparation tools, experiment tracking platforms, model training frameworks, and deployment monitoring tools.
At the same time, a new approach called neuro-symbolic AI (NeSy) and composite AI architectures is gaining traction. NeSy combines traditional machine learning with symbolic reasoning and knowledge graphs to make AI systems more explainable and easier to understand. The rapid advancement of generative AI and large language models (LLMs) has highlighted the critical importance of knowledge grounding in AI systems. Ensuring that model outputs are factual, interpretable, and verifiable is essential to maintaining the reliability and accountability of AI-driven decision-making.
The MLLO WG addresses these challenges by developing an ontology that provides: 1) ML framework-neutral standard representation, 2) abstract and dynamic binding capabilities, 3) increased FAIRness of ML models and datasets, 4) extensibility to accommodate evolving ML practices, and 5) integration with domain-specific knowledge encoded with other IOF ontology modules to bridge the gaps among teams and organizations.
Leadership
- Chair/Co-chair: Perawit Charoenwut, Hyunwoong Ko, Serm Kulvatunyou