arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2606.03232cs.LGcs.AI
收藏

GFFMERGE: 图神经力场的高效合并及其扩展

GFFMERGE: Efficient Merging of Graph Neural Force Fields and Beyond

University of California, Berkeley(加州大学伯克利分校);Stanford University(斯坦福大学);University of Cambridge(剑桥大学)

Parth Verma, Parv P. Singh, Vipul Garg, Ishita Thakre, N. M. Anoop Krishnan, Sayan Ranu

AI总结:

提出GFFMERGE框架,通过凸嵌入对齐问题解析解实现图神经网络的闭式模型合并,在力场回归任务中恢复接近联合训练的性能,并实现5-27倍加速。

摘要:

图神经网络(GNN)通过降低计算成本实现接近量子精度的原子模拟,彻底改变了神经力场,但将这些模型适应新化学系统需要对基础模型进行昂贵的重新训练。受视觉和语言处理中模型合并的启发,我们提出了GFFMERGE,这是第一个用于GNN闭式模型合并的原则性框架。我们利用消息传递层的线性结构,将合并问题形式化为具有解析解的凸嵌入对齐问题。通过对GNN模型合并的首次系统基准测试,我们发现为视觉和语言设计的现有方法在力场回归任务上灾难性地失败,而GFFMERGE恢复了接近黄金标准联合训练的性能。在分子(MD17、MD22)、固态(LiPS20)和大规模图基准测试中,GFFMERGE及其通用GNN对应物GNNMERGE实现了5-27倍的加速,同时支持专业模型的模块化组合。值得注意的是,我们的闭式解在微调前就优于所有基线方法,并为更快、数据高效的收敛提供了优越的初始化。

英文摘要:

Graph Neural Networks (GNNs) have revolutionized Neural Force Fields for atomistic simulations, achieving near-quantum accuracy at reduced cost, yet adapting these models to new chemical systems requires expensive retraining of foundation models. Inspired by model merging in vision and language processing, we introduce GFFMERGE, the first principled framework for closed-form model merging in GNNs. We exploit the linear structure of message-passing layers and formulate merging as a convex embedding-alignment problem with an analytical solution. Through the first systematic benchmarking of model merging for GNNs, we show that existing methods designed for vision and language catastrophically fail on force field regression, while GFFMERGE recovers performance approaching gold standard joint training. Across molecular (MD17, MD22), solid-state (LiPS20), and large-scale graph benchmarks, GFFMERGE and GNNMERGE (its generic GNN counterpart) achieve 5-27$\times$ speedups while enabling modular composition of specialized models. Remarkably, our closed-form solution alone outperforms all baseline methods before fine-tuning and provides superior initialization for faster, data-efficient convergence.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑