Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
The paper introduces Minor Component Adaptation (MiCA), a parameter-efficient fine-tuning method that adapts underutilized subspaces identified via SVD of model representations. By constraining parameter updates to minor singular directions, MiCA achieves up to 5.9x improvement in knowledge acquisition compared to LoRA. This approach allows for more efficient and stable integration of new knowledge with a smaller parameter footprint (6-60% of LoRA).
Forget LoRA, fine-tuning on the *least* important parameters can unlock up to 5.9x better knowledge acquisition in LLMs.
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.