Fig. 3From: Task-agnostic representation learning of multimodal twitter data for downstream applicationsImage retrieval task. A tweet’s components are projected in the joint embedding space along with images and the most similar image is retrievedBack to article page