Continual Learning for Generative Retrieval over Dynamic Corpora

By Jiangui Chen et al.

Published on Oct. 21, 2023

Read the original document by opening this link in a new tab.

Table of Contents

1. Introduction
2. Problem Statement
3. Methodology

Summary

This paper introduces a novel Continual-LEarner for generatiVE Retrieval (CLEVER) model that addresses the practical continual learning problem for generative retrieval (GR) over dynamic corpora. The model focuses on incrementally indexing new documents while preserving the ability to answer queries with both previously and newly indexed relevant documents. Key contributions include Incremental Product Quantization for low computational cost encoding of new documents and a memory-augmented learning mechanism to prevent forgetting of previously indexed documents. Experimental results demonstrate the effectiveness and efficiency of the proposed model.