Post by EDF STORE

🚀 STORE highlights an innovative contribution from Università degli Studi di Modena e Reggio Emilia: "Closed-Form Merging of Parameter-Efficient Modules for Federated Continual Learning". 🧠This research introduces LoRM, a new approach to merging multiple specialized AI models using low-rank adaptation (LoRA). Instead of simple averaging, LoRM uses an alternating optimization strategy to ensure that the merged model consistently reflects the behavior of all original modules, guaranteeing both scalability and performance. 🌍Applied to Federated Class-Incremental Learning, LoRM enables efficient integration of locally trained models, ensuring alignment across tasks and clients and achieving state-of-the-art results. 🤝As a STORE partner, Università degli Studi di Modena e Reggio Emilia continues to demonstrate its leadership in AI and deep learning. 👉 Watch the presentation & explore the code: https://lnkd.in/ePduVHy6   👉 Learn more about STORE: https://edf-store.com/ Riccardo Salami, Pietro Buzzega, Luigi Sabetta, Simone Calderara, Jacopo Bonato, Matteo Mosconi #STORE #AI #FederatedLearning #ContinualLearning #Innovation #UNIMORE