AI Paper Atlas

Research map / Graph learning / GNN architectures

GFFMERGE: Efficient Merging of Graph Neural Force Fields and Beyond

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

ICML 2026 · Graph learning › GNN architectures · approach: Equivariant networks for molecules · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
use closed-form embedding-alignment merging instead of weight averaging, Fisher, TIES or EMRM3GNet force fields · MD17 5-Task MixForce MAE: 2.16 -> 0.89Improved
use closed-form embedding-alignment merging instead of baselinesOrb force fields · MD17 5-Task MixForce MAE: 3.98 -> 0.94Improved
merge LiPS20 solid-state models with baseline merging methodsM3GNet · LiPS20Force MAE: baselines up to 51.79 vs GFFMerge 0.72 (beta + gamma Li3PS4)Hurt
add targeted fine-tuning of late interaction layers and force heads after closed-form merge (ablation)GFFMerge · LiPS20 Li3P + Li2SForce MAEImproved
closed-form initialization before any fine-tuning (epoch 0) compared to baselines (ablation)GFFMerge · held-out test setForce MAEImproved
fine-tune merged model with only 250 samples versus full data (ablation)GFFMerge · few-shot fine-tuningForce MAENo clear effect
use GFFMerge instead of Joint FT for trainingM3GNet and Orb · MD22 Ac-Ala3 + DHA + Stachyosetraining time speedup: 7358.42 s -> 264.62 s (27.81x)Improved
use GFFMerge instead of baselines in 1000-step MD rolloutsM3GNet · MD17 5-Task MixForce violation: 0.652 -> 0.562 (vs Wt Avg)Improved
merge more domains simultaneouslybaseline merging methods · MD17Force MAEHurt
use GNNMerge instead of Surgery for GCNs trained on disjoint label splitsGCN · in-domain node classificationaccuracy: 78.01 -> 88.94 (average)Improved
merge GCNs of different tasks (node classification and link prediction) with baselinesGCN · different tasks on different datasetslink prediction AUCHurt
merge MAG240M and Papers100M GraphSAGE models instead of joint trainingGraphSAGE · OGB large scaletime and memory: 650GB RAM -> ~2 GB, under 8 minutesImproved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy