FLORA is a novel machine learning approach that leverages federated learning and parameter-efficient adapters to train visual-language models (VLMs) in distributed environments.
FLORA utilizes Low-Rank Adaptation (LoRA) in combination with Federated Learning to fine-tune VLMs like the CLIP model, preserving data privacy and minimizing communication overhead.
Experimental evaluations show FLORA's superior accuracy, efficiency, and adaptability compared to traditional federated learning methods in both IID and non-IID settings.
FLORA's efficient model adaptation addresses real-world data challenges in distributed learning environments, making it a promising solution.