This framework was developed by Sun Yat-sen University and the Chinese University of Hong Kong in their paper 「DQ-LORE: DUAL QUERIES WITH LOW RANK APPROXIMATION RE-RANKING FOR IN-CONTEXT LEARNING」Proposed in.

In this study, the team introduced a framework that uses "dual query (DQ) and low-rank approximate rearrangement (LoRe)" Automatically select contextual learning examples. Experiments show that DQ-LoRe surpasses previous methods in automatically selecting GPT-4 examples, with an accuracy improvement from 92.5% to 94.2%, opening a new path for LLMs to solve complex reasoning problems. The research team's comprehensive analysis further shows that DQ-LoRe consistently outperforms retrieval-based methods in performance and adaptability, especially in scenarios characterized by distribution changes. DQ-LoRe pushes the boundaries of contextual learning and opens new paths to solve complex reasoning challenges.