Privacy-preserving continuous learning for MobileSAM via Federated Learning
Ji Su Yoon, Yu Min Park, Chaoning Zhang, Choong Seon Hong · 2023
The latest foundation model is the most notable technology in the field of artificial intelligence. The Segment Anything Model (SAM), which is currently receiving tremendous attention, is one of the foundation models that will bring about a breakthrough in the field of image segmentation. Federated learning is a very suitable learning structure for efficiently learning these foundation models. Therefore, developing federated learning structures for foundation models is one of the important objectives in the field of federated learning in recent years. However, learning the current foundation model through a federated learning structure is extremely difficult. In particular, the complexity of the model is so high that it requires a very high level of computing resources there are many limitations to doing this on local devices. Therefore, we use a foundational model for image segmentation tasks through the recently published MobileSAM. MobileSAM is a lightweight version of the SAM that can also be learned in federated learning structures. In this paper, we propose a federated learning structure with MobileSAM for privacy-preserving continuous learning. Experiments have shown that MobileSAM learned from federated learning has sufficiently available performance.