Adaptive Federated Knowledge Injection into Medical Foundation Models

Explore cutting-edge research on integrating medical knowledge into foundation models while preserving data privacy in this Stanford University MedAI Group Exchange Session. Learn about the Federated Medical Knowledge Injection (FEDMEKI) platform, a novel benchmark designed to address the unique challenges of developing comprehensive medical foundation models under stringent privacy regulations. Discover the innovative FedKIM approach, which leverages lightweight local models to extract healthcare knowledge from private data and integrates this knowledge into centralized foundation models using an adaptive Multitask Multimodal Mixture Of Experts (M3OE) module. Examine how this methodology enables medical foundation models to learn from diverse medical knowledge without direct data exposure, simultaneously preserving privacy while enhancing the model's ability to handle complex medical tasks across multiple modalities. Review extensive experimental results across twelve tasks in seven modalities that demonstrate FedKIM's effectiveness in various settings and its potential to scale medical foundation models without accessing sensitive data. Gain insights from Xiaochen Wang, a Ph.D. candidate at Penn State University specializing in healthcare informatics, multimodal learning, and information retrieval, whose research has been recognized at top-tier AI conferences including ACL, EMNLP, NAACL, NeurIPS, and KDD.