SOURCE-LINKED INTELLIGENCE

Two-Stage Mixture-of-LoRA for Multi-Task Medical Vision-Language Learning

arXiv · AI, language, vision and robotics · article · Sep 13, 2026 · UTC

Medical vision-language models (VLMs) allow a single model to perform clinical image analysis tasks ranging from diagnosis classification to report generation. However, joint adaptation is challenged by heterogeneous output formats, conflicting task gradients, and imbalanced training data. Hence, we present \textbf{Two-Stage Mixture-of-LoRA}, a framework built on MedGemma-1.5-4B. The framework uses a shared-specific Mixture-of-LoRA architecture comprising one shared LoRA and six task-specific expert LoRAs, together with a two-stage training procedure. In Stage 1, we jointly train the shared Lo

Read original source ↗ Open in workspace

recordType

paper

region

Global