Papers
arxiv:2604.17078

Understanding and Enforcing Weight Disentanglement in Task Arithmetic

Published on Apr 18
ยท Submitted by
Shangge Liu
on Apr 22
Authors:
,
,
,
,
,
,

Abstract

Task arithmetic lacks theoretical explanation for its success, but the proposed OrthoReg method addresses this by promoting weight disentanglement through enforced orthogonality in weight updates during fine-tuning.

AI-generated summary

Task arithmetic provides an efficient, training-free way to edit pre-trained models, yet lacks a fundamental theoretical explanation for its success. The existing concept of ``weight disentanglement" describes the ideal outcome of non-interfering task composition but does not reveal its underlying cause. Crucially, what intrinsic properties of the pre-trained model (ฮธ_0) or the task vectors (ฯ„_t) enable this disentanglement remains underexplored. In this paper, we introduce Task-Feature Specialization (TFS), a model's ability to allocate distinct internal features to different tasks, as the fundamental principle. We first prove that TFS is a sufficient condition for weight disentanglement. More importantly, we find that TFS also gives rise to an observable geometric consequence: weight vector orthogonality. This positions TFS as the common cause for both the desired functional outcome (disentanglement) and a measurable geometric property (orthogonality). This relationship provides the key insight for our method: since the abstract TFS property is intractable to enforce directly, we can instead promote weight disentanglement by shaping its concrete geometric consequence, orthogonality. Therefore, we propose OrthoReg, a simple and effective regularization method that actively enforces an internal orthogonal structure on weight updates (ฮ”W) that constitute ฯ„_t during fine-tuning. And we theoretically prove that OrthoReg promotes disentanglement. Extensive experiments demonstrate that OrthoReg consistently and significantly enhances the performance of various task arithmetic methods. Code is available at https://github.com/RL-MIND/OrthoReg{https://github.com/RL-MIND/OrthoReg}.

Community

Paper author Paper submitter

Hi everyone! ๐Ÿค— We are thrilled to share our latest work accepted at CVPR 2026: "Understanding and Enforcing Weight Disentanglement in Task Arithmetic".

If you are interested in Model Merging and Task Arithmetic, here is a quick overview of what we did and how you can use it!

๐Ÿ’ก The Core Problem: Why does Task Arithmetic work?
While task arithmetic (adding/subtracting task vectors) is an elegant, training-free way to edit models, its underlying mechanism has remained unclear. We theoretically prove that Task-Feature Specialization (TFS) is a sufficient condition for weight disentanglement. More importantly, TFS naturally leads to an observable geometric property: Weight Vector Orthogonality.

๐Ÿ› ๏ธ Our Solution: OrthoReg
Since enforcing abstract feature specialization is intractable, we propose OrthoRegโ€”a simple, plug-and-play regularization term added to the standard fine-tuning loss. It actively enforces column-wise orthogonality on the weight updates (\Delta W) during fine-tuning.

The loss is incredibly simple to implement:
L=Ltask+ฮปโ‹…โˆ‘lโˆฅ(ฮ”W(l))โŠคฮ”W(l)โˆ’IโˆฅF2\mathcal{L} = \mathcal{L}_{\text{task}} + \lambda \cdot \sum_l \left\|(\Delta W^{(l)})^\top \Delta W^{(l)} - I\right\|_F^2

๐Ÿ“Š Key Results

  • Consistent Boosts: Significantly improves the performance of various task arithmetic baselines (Non-linear FT, TTA, ATT-FT, LoRA) across ViT-B-32, ViT-B-16, and ViT-L-14.
  • High Efficiency: Achieves the disentanglement benefits of Tangent Task Arithmetic (TTA) but without the massive computational overhead of Jacobian calculations.
  • Cleaner Forgetting: Substantially improves task negation (making a model forget a specific task) with minimal side effects on general capabilities.

๐Ÿ”— Resources
We believe in open science and have released all our artifacts:

We would love to hear your thoughts, feedback, or answer any questions you might have. Feel free to drop a comment below! ๐Ÿ‘‡

Sign up or log in to comment

Get this paper in your agent:

hf papers read 2604.17078
Don't have the latest CLI?
curl -LsSf https://hf.co/cli/install.sh | bash

Models citing this paper 2

Datasets citing this paper 0

No dataset linking this paper

Cite arxiv.org/abs/2604.17078 in a dataset README.md to link it from this page.

Spaces citing this paper 0

No Space linking this paper

Cite arxiv.org/abs/2604.17078 in a Space README.md to link it from this page.

Collections including this paper 0

No Collection including this paper

Add this paper to a collection to link it from this page.