GASE: Generatively Augmented Sentence Encoding

Manuel Frank, Haithem Afli

We propose a training-free approach to improve sentence embeddings leveraging test-time compute by applying generative text models for data augmentation at inference time. Unlike c…