Continual Model Merging without Data: Dual Projections for Balancing Stability and Plasticity. NeurIPS, 2025.

Abstract

Model merging integrates multiple expert models with diverse capabilities into a unified framework, facilitating collaborative learning. However, most existing methods assume simultaneous access to all models, which is often impractical in real-world scenarios where models are received sequentially. While some studies have investigated continual model merging (CMM)--which involves sequentially merging multiple models--the challenge of balancing prior knowledge (stability) and incorporating new tasks (plasticity) remains unresolved. This paper, for the first time, formally defines the stability and plasticity of CMM from the perspective of orthogonal projection. Subsequently, we analyze the relationships among the spaces spanned by task data, historical gradients, and accumulated gradients. Building on this, we propose a data-free Dual Orthogonal Projection (DOP) method, which eliminates data dependence and mitigates interference between the merged model and models for old and new tasks by projecting their parameter differences onto their respective approximate data spaces. Finally, to solve potential conflicts between stability and plasticity, we reformulate DOP as a multi-objective optimization problem and employ a multi-gradient descent algorithm to obtain a Pareto-optimal solution. Extensive experiments across multiple architectures and task configurations validate that our approach significantly outperforms state-of-the-art CMM methods.

Citation

If you find our paper or this resource helpful, please consider cite:

@article{DOP_NeurIPS2025,
  title={Continual Model Merging without Data: Dual Projections for Balancing Stability and Plasticity},
  author={Yang, Enneng and Tang, Anke and Shen, Li and Guo, Guibing and Wang, Xingwei and Cao, Xiaochun and Zhang, Jie},
  journal={The Thirty-ninth Annual Conference on Neural Information Processing Systems},
  year={2025}
}

Thanks!

Development environment configuration

The project relies on the FusionBench-v0.2.9, which can be reference for the library to configure the basic environment, or in accordance with the following steps to create a development environment:

Step 1: Create a Conda environment

conda create --name dop python=3.12.4

Step 2: Activate the Conda environment

conda activate dop

Step 3: Install the dependent environment for the project

git clone https://github.com/EnnengYang/DOP
cd DOP

pip install -e . # install the package in editable mode

Datasets and Checkpoints

When executing the script, the required files will be automatically downloaded from Huggingface. Please ensure that you can access Huggingface.

Run the experiment

Ref 'examples/continual_dop'

bash main_vitb32.sh

bash main_vitb16.sh

bash main_vitl14.sh

Acknowledgements

This project is based on FusionBench.

Our code has been integrated into FusionBench.