Unsupervised Corpus Aware Language Model Pre-training for Dense Passage Retrieval
This addresses the challenge of training dense retrievers more efficiently for information retrieval tasks, though it is incremental as it builds on the existing Condenser architecture.
The paper tackled the problems of dense retrievers being fragile to training data noise and requiring large batches for robust embedding space learning by proposing coCondenser, which adds an unsupervised corpus-level contrastive loss to the Condenser pre-training architecture. It achieved comparable performance to the state-of-the-art RocketQA system on MS-MARCO, Natural Question, and Trivia QA datasets without needing heavy data engineering or large batch training.
Recent research demonstrates the effectiveness of using fine-tuned language models~(LM) for dense retrieval. However, dense retrievers are hard to train, typically requiring heavily engineered fine-tuning pipelines to realize their full potential. In this paper, we identify and address two underlying problems of dense retrievers: i)~fragility to training data noise and ii)~requiring large batches to robustly learn the embedding space. We use the recently proposed Condenser pre-training architecture, which learns to condense information into the dense vector through LM pre-training. On top of it, we propose coCondenser, which adds an unsupervised corpus-level contrastive loss to warm up the passage embedding space. Retrieval experiments on MS-MARCO, Natural Question, and Trivia QA datasets show that coCondenser removes the need for heavy data engineering such as augmentation, synthesis, or filtering, as well as the need for large batch training. It shows comparable performance to RocketQA, a state-of-the-art, heavily engineered system, using simple small batch fine-tuning.