AI Paper Atlas

Research map / Architectures and efficiency / RNNs and normalization

How does the optimizer implicitly bias the model merging loss landscape?

Chenxiang Zhang, Alexander Theus, Damien Teney, Antonio Orvieto, Jun Pang, Sjouke Mauw

ICLR 2026 · Architectures and efficiency › RNNs and normalization · approach: Layer-wise analysis of language models

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
increase learning rate (2e-1 vs 1e-2) at similar single-model accuracy (ablation)ResNet18 SGD · CIFAR100median merge gain (%): +0.2% -> +1.2%Improved
use too large a learning rate (ablation)ResNet18 / DenseNet121 SGD · CIFAR10, CIFAR100, TinyImageNetmerging successHurt
increase weight decay (scale-invariant nets) (ablation)DenseNet121 · TinyImageNetmedian merge gain (%): +0.5% -> +1.2%Improved
vary weight decay in non-scale-invariant MLP (ablation)MLP · SVHNmerge gainNo clear effect
decrease batch size from 256 to 16 (ablation)ResNet18 · CIFAR100median merge gain (%): ~0% -> +1%Improved
increase momentum to 0.9 (ablation)ResNet18 · CIFAR100median merge gain (%): +0.2% -> +1.0%Improved
add data augmentation (ablation)ResNet18 · CIFAR100merge gainImproved
increase learning rate (ablation)CLIP ViT-B/16 transfer · WILDS-FMoWaccuracy gain of merged model: 4x larger gain at 1e-4 than 1e-6Improved
use largest learning rate for finetuning (ablation)CLIP ViT-B/16 transfer · WILDS-FMoWmerged model accuracyHurt
use larger learning rate with non-pretrained initialization (ablation)ConvNext-T task arithmetic · CIFAR10 non-transferloss landscape flatness under alpha-interpolationHurt
use TIES merging instead of task arithmetic (ablation)CLIP ViT-B/16 · FMoW+RESISC45 at lr 3e-5normalized accuracy (%): 85.9 -> 88.0Improved
merge models trained with different learning rates (ablation)CLIP ViT-B/16 · FMoW+RESISC45merged accuracyHurt

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