MGDT: MLLM-Guided Diffusion Transformer with Relation-Adaptive Mixture-of-Experts for Multimodal Knowledge Graph Completion

Explain Like I'm Normal

Researchers have developed a new framework called MGDT that uses Multi-modal Large Language Models to guide Diffusion Transformers for filling in missing data in knowledge graphs. By using a Relation-Adaptive Mixture-of-Experts, the system separates semantic alignment from entity generation, resulting in much cleaner data recovery. This 'align-then-diffuse' approach prevents the messy noise typical of processing raw multimodal features simultaneously.