AI Paper Atlas

Research map / Architectures and efficiency / Pruning and efficient tuning

FW-Merging: Scaling Model Merging with Frank-Wolfe Optimization

Hao Mark Chen, Shell Xu Hu, Wayne Luk, Timothy Hospedales, Hongxiang Fan

ICCV 2025 · Architectures and efficiency › Pruning and efficient tuning · approach: Low-rank compression and merging · 2 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 layer-wise LMO instead of task-wise (ablation)FW-TA · NLP discriminative tasksavg score: 78.2 -> 85.4Improved
use layer-wise LMO instead of task-wise (ablation)FW-TA · vision 4 CV tasksavg accuracy: 73.7 -> 74.0Improved
use task-wise LMO instead of layer-wise (ablation)FW-AM · vision 4 CV tasksavg accuracy: 79.7 -> 80.1Improved
optimize merging coefficients lambda (ablation)FW-AM · vision 4 CV tasksavg accuracy: 70.3 -> 80.1Improved
use soft FW (multiple top-k vertices) instead of hard FW (ablation)FW-AM vs FW-TA · vision merging many modelsavg accuracy: up to 6.7 pointsImproved
initialize FW-Merging from Adamerging instead of pre-trained model (ablation)FW-Merging · vision benchmarkaccuracyImproved
initialize FW-Merging from Task Arithmetic instead of pre-trained model (ablation)FW-Merging · vision benchmarkaccuracyHurt
apply FW-Merging on top of Ties-Merging as merging function (ablation)FW-Merging · NLP and visionaccuracyImproved
add 16 irrelevant models to merging poolFW-AM · 4 CV tasksavg accuracy: 73.9 to 74.1No clear effect
add irrelevant models to merging poolTask Arithmetic · 4 CV tasksavg accuracy: 49.1% dropHurt
add irrelevant models to merging poolDARE · 4 CV tasksavg accuracy: 64.4% dropHurt
add relevant models to merging poolFW-AM · 20 CV tasksavg accuracy: 59.2 -> 68.3Improved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

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