Energy-Efficient and Privacy-Preserving Federated Continual Learning for Cultural Heritage Preservation and Digital Humanities

Abstract - Cultural heritage institutions increasingly adopt AI to process complex, distributed datasets like digitized artifacts and historical records. However, traditional AI approaches raise critical concerns regarding high energy consumption, privacy risks to sensitive data, and an inability to adapt to evolving collections. This paper surveys Federated Continual Learning (FCL) as a sustainable and ethically responsible solution that enables collaborative AI across institutions without centralizing sensitive data. We show how FCL reduces the computational footprint through distributed processing and efficient protocols, aligning with Green AI principles. FCL incorporates strong privacy guarantees like differential privacy and secure aggregation, preserving cultural asset integrity while enabling knowledge sharing. We present a conceptual framework with implementation strategies, identifying challenges and future research directions. Finally, a case study of a state-of-the-art FCL method illustrates its application for trustworthy, energy-efficient artifact classification in digital humanities.