In-context Learning with Retrieved Demonstrations for Language Models: A Survey

TL;DR

This survey reviews recent advancements in in-context learning with retrieved demonstrations, highlighting how retrieval-based methods improve adaptability, efficiency, and reduce biases in large language models.

Contribution

It provides a comprehensive comparison of design choices, retrieval models, training procedures, and inference algorithms in the emerging area of retrieval-augmented in-context learning.

Findings

Retrieval-based demonstrations enhance model adaptability and scalability.

Using retrieved demonstrations reduces biases compared to manual selection.

The survey identifies key design choices and challenges in retrieval-augmented ICL.

Abstract

Language models, especially pre-trained large language models, have showcased remarkable abilities as few-shot in-context learners (ICL), adept at adapting to new tasks with just a few demonstrations in the input context. However, the model's ability to perform ICL is sensitive to the choice of the few-shot demonstrations. Instead of using a fixed set of demonstrations, one recent development is to retrieve demonstrations tailored to each input query. The implementation of demonstration retrieval is relatively straightforward, leveraging existing databases and retrieval systems. This not only improves the efficiency and scalability of the learning process but also has been shown to reduce biases inherent in manual example selection. In light of the encouraging results and growing research in ICL with retrieved demonstrations, we conduct an extensive review of studies in this area. In this survey, we discuss and compare different design choices for retrieval models, retrieval training procedures, and inference algorithms.