Reduce Catastrophic Forgetting of Dense Retrieval Training with Teleportation Negatives

TL;DR

This paper introduces ANCE-Tele, a method that reduces catastrophic forgetting in dense retrieval training by using momentum and lookahead negatives, leading to more stable training and better performance.

Contribution

The paper proposes ANCE-Tele, a novel negative sampling technique that mitigates catastrophic forgetting and improves dense retrieval training stability and effectiveness.

Findings

ANCE-Tele outperforms previous state-of-the-art systems.

It eliminates the dependency on sparse negatives.

It is competitive with larger models (50x parameters).

Abstract

In this paper, we investigate the instability in the standard dense retrieval training, which iterates between model training and hard negative selection using the being-trained model. We show the catastrophic forgetting phenomena behind the training instability, where models learn and forget different negative groups during training iterations. We then propose ANCE-Tele, which accumulates momentum negatives from past iterations and approximates future iterations using lookahead negatives, as "teleportations" along the time axis to smooth the learning process. On web search and OpenQA, ANCE-Tele outperforms previous state-of-the-art systems of similar size, eliminates the dependency on sparse retrieval negatives, and is competitive among systems using significantly more (50x) parameters. Our analysis demonstrates that teleportation negatives reduce catastrophic forgetting and improve convergence speed for dense retrieval training. Our code is available at https://github.com/OpenMatch/ANCE-Tele.