Asymmetry in Low-Rank Adapters of Foundation Models
This work addresses parameter-efficient fine-tuning for large AI models, offering insights that could reduce computational costs, though it is incremental in nature.
The paper investigates the distinct roles of matrices in Low-Rank Adaptation (LoRA) for fine-tuning foundation models, finding that fine-tuning matrix B is more effective than matrix A, and a random untrained A performs nearly as well as a fine-tuned one, with experiments on models like RoBERTa and LLaMA-2.
Parameter-efficient fine-tuning optimizes large, pre-trained foundation models by updating a subset of parameters; in this class, Low-Rank Adaptation (LoRA) is particularly effective. Inspired by an effort to investigate the different roles of LoRA matrices during fine-tuning, this paper characterizes and leverages unexpected asymmetry in the importance of low-rank adapter matrices. Specifically, when updating the parameter matrices of a neural network by adding a product $BA$, we observe that the $B$ and $A$ matrices have distinct functions: $A$ extracts features from the input, while $B$ uses these features to create the desired output. Based on this observation, we demonstrate that fine-tuning $B$ is inherently more effective than fine-tuning $A$, and that a random untrained $A$ should perform nearly as well as a fine-tuned one. Using an information-theoretic lens, we also bound the generalization of low-rank adapters, showing that the parameter savings of exclusively training $B$ improves the bound. We support our conclusions with experiments on RoBERTa, BART-Large, LLaMA-2, and ViTs.