MiCA Learns More Knowledge Than LoRA and Full Fine-Tuning

TL;DR

MiCA is a new parameter-efficient fine-tuning method that focuses on minor subspaces of model representations, outperforming LoRA in knowledge acquisition with fewer parameters.

Contribution

Introduces MiCA, a novel fine-tuning approach targeting minor singular vectors, enhancing knowledge learning efficiency over existing methods like LoRA.

Findings

MiCA achieves up to 5.9x better knowledge acquisition.

MiCA uses only 6-60% of parameters compared to LoRA.

MiCA provides a more stable fine-tuning mechanism.

Abstract

Minor Component Adaptation (MiCA) is a novel parameter-efficient fine-tuning method for large language models that focuses on adapting underutilized subspaces of model representations. Unlike conventional methods such as Low-Rank Adaptation (LoRA), which target dominant subspaces, MiCA leverages Singular Value Decomposition to identify subspaces related to minor singular vectors associated with the least significant singular values and constrains the update of parameters during fine-tuning to those directions. This strategy leads to up to 5.9x improvement in knowledge acquisition under optimized training hyperparameters and a minimal parameter footprint of 6-60% compared to LoRA. These results suggest that constraining adaptation to minor singular directions provides a more efficient and stable mechanism for integrating new knowledge into pre-trained language models.