The author introduces DQ-LoRe, a framework leveraging Dual Queries and Low-rank approximation Re-ranking to enhance in-context learning. By selecting exemplars based on CoT and employing PCA for dimensionality reduction, DQ-LoRe outperforms existing methods in multi-step reasoning tasks.
自動選択のためのDual QueriesとLow-rank近似再ランキングを活用するフレームワーク、DQ-LoReは、GPT-4の例示物の自動選択において優れたパフォーマンスを発揮します。
DQ-LoRe significantly enhances exemplar selection for in-context learning, outperforming existing methods.