REALM at CAiSE Conference

REALM project partners Ankur Lohachab and Visara Urovi from Maastricht University wrote a publication titled: "A Blockchain-Based Approach for Model Card Accountability and Regulatory Compliance" on managing and sharing model cards using blockchain technology.

Last week, Ankur Lohachab presented their work at the 36th International Conference on Advanced Information Systems Engineering CAiSE Conference 2024 in Limassol, Cyprus.

Supported by REALM, Ankur and Visara's work explores various aspects, including making interactions among healthcare stakeholders seamless while ensuring robust role-based access control.

Paper abstract

This paper introduces an approach that utilizes smart contracts to facilitate the trustworthy sharing and management of Machine Learning (ML) and Artificial Intelligence (AI) models, as described using model cards. To this end, the proposed approach incorporates Account Abstraction for authentication, enabling role-based access control. This control allows stakeholders to share, track, and validate model cards transparently and securely while tailoring visibility and interaction to preserve privacy in accordance with each role’s privileges. The approach further delineates the conceptualization and lifecycle management of model cards, spanning from creation to deprecation, all within a blockchain-based framework. Additionally, the paper discusses the state parameterization of model cards, formalizing the operational dynamics and constraints associated with each phase of their progression. The proof of concept, implemented to evaluate our approach, suggests that it is capable of effectively capturing and maintaining an immutable record of the various states of model cards, thereby providing a robust and verifiable trail. Overall, our approach is designed to ensure the integrity of model cards and establish accountability, thereby strengthening trust among stakeholders, particularly those relying on AI and ML models as described in model cards.

Read the whole paper: https://link.springer.com/chapter/10.1007/978-3-031-61003-5_4