AI Paper Atlas

Research map / Architectures and efficiency / Mixture of experts

DuetMerging: Synergizing Dynamic and Static Strategies for Mitigating Task Interference in Model Merging

Yan Li, Guiping Cao, Yaguang Song, Ming Tao, Haoran Gong, Junhui Liu, Yaowei Wang, Dongmei Jiang

CVPR 2026 · Architectures and efficiency › Mixture of experts · approach: Expert routing and specialization

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
add neuron-guided static correction on top of Tucker dynamic merging (ablation)DuetMerging · 8-task average, ViT-B/32average accuracy: 89.2 -> 89.8Improved
add neuron-guided static correction (ablation)DuetMerging · 8-task average, ViT-L/14average accuracy: 93.3 -> 93.8Improved
use Tucker decomposition instead of per-task SVD for the expert pool (ablation)DuetMerging · 8-task average, ViT-B/32average accuracy: 87.7 -> 89.2Improved
use Tucker decomposition instead of per-task SVD for the expert pool (ablation)DuetMerging · 8-task average, ViT-L/14average accuracy: 92.1 -> 93.3Improved
add neuron sparsification to Task ArithmeticTask Arithmetic · 8-task average, ViT-B/32average accuracy: 69.0 -> 75.5Improved
add neuron sparsification to TIES-MergingTIES-Merging · 8-task average, ViT-B/32average accuracy: 72.8 -> 75.1Improved
add neuron sparsification to AdaMergingAdaMerging · 8-task average, ViT-B/32average accuracy: 80.1 -> 80.5Improved
DuetMerging versus fully supervised Multi-Task modelDuetMerging · 8-task average, ViT-B/32average normalized accuracy: 98.3 -> 99.2Improved
DuetMerging versus AdaMerging on impulse noise corruptionDuetMerging · ImageNet-C Impulse Noiseaccuracy: 68.0 -> 67.6Hurt

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