Decomposing Task Vectors for Refined Model Editing
This addresses the challenge of precise control in model editing for AI practitioners, though it is incremental as it builds on existing task vector methods.
The paper tackled the problem of overlapping concepts in task vectors interfering during arithmetic operations for model editing, and the result was a decomposition method that improved multi-task merging by 5% in image classification, enabled clean style mixing in diffusion models, and reduced toxicity by 47% in language models while preserving performance.
Large pre-trained models have transformed machine learning, yet adapting these models effectively to exhibit precise, concept-specific behaviors remains a significant challenge. Task vectors, defined as the difference between fine-tuned and pre-trained model parameters, provide a mechanism for steering neural networks toward desired behaviors. This has given rise to large repositories dedicated to task vectors tailored for specific behaviors. The arithmetic operation of these task vectors allows for the seamless combination of desired behaviors without the need for large datasets. However, these vectors often contain overlapping concepts that can interfere with each other during arithmetic operations, leading to unpredictable outcomes. We propose a principled decomposition method that separates each task vector into two components: one capturing shared knowledge across multiple task vectors, and another isolating information unique to each specific task. By identifying invariant subspaces across projections, our approach enables more precise control over concept manipulation without unintended amplification or diminution of other behaviors. We demonstrate the effectiveness of our decomposition method across three domains: improving multi-task merging in image classification by 5% using shared components as additional task vectors, enabling clean style mixing in diffusion models without generation degradation by mixing only the unique components, and achieving 47% toxicity reduction in language models while preserving performance on general knowledge tasks by negating the toxic information isolated to the unique component. Our approach provides a new framework for understanding and controlling task vector arithmetic, addressing fundamental limitations in model editing operations.