Retrieval Augmentation

Desmond Elliott

Assistant Professor

Rita Ramos

Ph.D Fellow

(co-advised with Bruno Martins)

Retrieval augmentation has proven a useful approach to improving the quality and relevance of generated texts using in-context examples, as opposed to relying on what can be learned and store in model weights. We have worked on applying retrieval augmentation in a multimodal setting, with multimodal encoders, independent pretrained models, or in a multilingual setting without any supervised training.