Multimodal learning with graphs

Artificial intelligence for graphs (graph AI) has achieved remarkable success in modeling complex systems, ranging from dynamic networks in biology to interacting particle systems in physics. However, the increasingly heterogeneous graph datasets call for multimodal methods that can combine different inductive biases: the set of assumptions that algorithms use to make predictions for inputs they have not encountered during training. Learning on multimodal graph datasets presents fundamental challenges because the inductive biases can vary by data modality and graphs might not be explicitly given in the input. To address these challenges, multimodal graph AI methods combine different modalities while leveraging cross-modal dependencies. Here, we survey 145 studies in graph AI and realize that diverse datasets are increasingly combined using graphs and fed into sophisticated multimodal methods, specified as image-intensive, knowledge-grounded and language-intensive models. Using this categorization, we introduce a blueprint for multimodal graph AI to study existing methods and guide the design of future methods.