Summary of Multi-task Model Merging Via Adaptive Weight Disentanglement, by Feng Xiong et al.
Multi-Task Model Merging via Adaptive Weight Disentanglement
by Feng Xiong, Runxi Cheng, Wang Chen, Zhanqiu Zhang, Yiwen Guo, Chun Yuan, Ruifeng Xu
First submitted to arxiv on: 27 Nov 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Computation and Language (cs.CL); Computer Vision and Pattern Recognition (cs.CV)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary This paper proposes an Adaptive Weight Disentanglement (AWD) method to tackle interference among tasks in model merging. The approach is motivated by the insight that task vectors should be orthogonal to minimize conflict. AWD initializes redundant vectors, subtracts them from original task vectors, and applies a norm constraint to preserve performance. Experimental results show AWD effectively extracts redundant vectors, retains robust performance, and achieves superior fusion outcomes. The paper demonstrates its technique on Task Arithmetic (TA) and other tasks, showcasing the potential for improved model performance. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Researchers have found a way to make AI models work better by combining information from different tasks. They used an idea called “model merging” where they added or subtracted weights from different tasks. The problem is that these weights can interfere with each other, making the model worse. To fix this, the researchers developed a new method called Adaptive Weight Disentanglement (AWD). AWD makes sure the weights don’t interfere by subtracting redundant information and keeping the important parts. The results show that AWD works really well and can make AI models more powerful. |
