Model-Agnostic Framework for Continuous Learning in Frozen Medical VLMs

This framework enables frozen multimodal models to incorporate new clinical evidence through three external mechanisms: a Skill for reasoning, a Knowledge Memory for facts, and a Multimodal Knowledge Base for visual examples. It bypasses the need for weight updates or fine-tuning to prevent medical knowledge obsolescence.