一作ACL 2026Findings· Findings of ACL
Polynomial Expansion Rank Adaptation: Enhancing Low-Rank Fine-Tuning with High-Order Interactions
Wenhao Zhang · Lin Mu · Li Ni · Peiquan Jin · Yiwen Zhang
摘要
Low-rank adaptation (LoRA) is a widely used strategy for efficient fine-tuning of large language models (LLMs), but its strictly linear structure fundamentally limits expressive capacity. The bilinear formulation of weight updates captures only first-order dependencies between low-rank factors, r...