Improving Retrieval-Augmented Large Language Models via Data Importance Learning
This work addresses performance bottlenecks in retrieval-augmented LLMs for tasks like question answering, offering a practical improvement but is incremental as it builds on existing retrieval augmentation methods.
The paper tackles the problem of improving retrieval-augmented large language models by addressing data quality limitations in retrieval corpora, proposing algorithms to compute data importance for pruning or reweighting, resulting in enhanced performance where small models with retrieval can outperform GPT-3.5 on some tasks.
Retrieval augmentation enables large language models to take advantage of external knowledge, for example on tasks like question answering and data imputation. However, the performance of such retrieval-augmented models is limited by the data quality of their underlying retrieval corpus. In this paper, we propose an algorithm based on multilinear extension for evaluating the data importance of retrieved data points. There are exponentially many terms in the multilinear extension, and one key contribution of this paper is a polynomial time algorithm that computes exactly, given a retrieval-augmented model with an additive utility function and a validation set, the data importance of data points in the retrieval corpus using the multilinear extension of the model's utility function. We further proposed an even more efficient (ε, δ)-approximation algorithm. Our experimental results illustrate that we can enhance the performance of large language models by only pruning or reweighting the retrieval corpus, without requiring further training. For some tasks, this even allows a small model (e.g., GPT-JT), augmented with a search engine API, to outperform GPT-3.5 (without retrieval augmentation). Moreover, we show that weights based on multilinear extension can be computed efficiently in practice (e.g., in less than ten minutes for a corpus with 100 million elements).