Lluna Uses Graph-Based Collaboration for Missing-Modality Inference

Lluna addresses accuracy drops in multimodal models when inputs are missing by combining direct unimodal prediction paths with a common graph-based collaboration branch. This architecture provides residual corrections to maintain performance even when specific modalities are unavailable.