The paper "NATiVE: Multi-modal Knowledge Graph Completion in the Wild" addresses the challenges of multi-modal knowledge graph completion (MMKGC) in real-world scenarios, where modality information is diverse and imbalanced. The authors propose a comprehensive framework called NATiVE to tackle these issues. NATiVE includes two key modules: Relation-guided Dual Adaptive Fusion (ReDAF) and Collaborative Modality Adversarial Training (CoMAT). ReDAF enables adaptive fusion of any modalities by considering relational context, while CoMAT uses adversarial training to augment imbalanced modality information. The authors construct a new benchmark called WildKGC with five datasets to evaluate their method, achieving state-of-the-art performance across different datasets and scenarios. The paper also provides theoretical analysis and comprehensive experiments to demonstrate the effectiveness, efficiency, and generalizability of NATiVE.The paper "NATiVE: Multi-modal Knowledge Graph Completion in the Wild" addresses the challenges of multi-modal knowledge graph completion (MMKGC) in real-world scenarios, where modality information is diverse and imbalanced. The authors propose a comprehensive framework called NATiVE to tackle these issues. NATiVE includes two key modules: Relation-guided Dual Adaptive Fusion (ReDAF) and Collaborative Modality Adversarial Training (CoMAT). ReDAF enables adaptive fusion of any modalities by considering relational context, while CoMAT uses adversarial training to augment imbalanced modality information. The authors construct a new benchmark called WildKGC with five datasets to evaluate their method, achieving state-of-the-art performance across different datasets and scenarios. The paper also provides theoretical analysis and comprehensive experiments to demonstrate the effectiveness, efficiency, and generalizability of NATiVE.