Energy Aware Generative AI Model Deployment

• Entity: Fundación Ramon Areces (Ayudas senior de investigación en ciencias sociales)
• Duration: 2025 - 2027
• Principal Investigator: Jordi Nin
• Participants: ESADE    

This project investigates how differential replication, or model copying, can make the adaptation of generative AI systems more energy-efficient. Instead of repeatedly retraining large models from scratch, the project develops methods to transfer their knowledge into tailored model copies that meet evolving business, technical, and regulatory requirements while reducing computational costs. The research focuses on three key challenges: enabling incremental learning as new data arrive, extending model-copying techniques to multimodal data such as text and images, and supporting compliance with the right to be forgotten by efficiently removing the influence of personal data. Through theoretical advances, real-world experiments, open-source software, and international collaboration, the project aims to promote more sustainable, adaptable, and privacy-conscious AI deployment.