AI Paper Atlas

Research map / Architectures and efficiency / Pruning and efficient tuning

Preference-Aligned LoRA Merging: Preserving Subspace Coverage and Addressing Directional Anisotropy

Wooseong Jeong, Wonyoung Lee, Kuk-Jin Yoon

CVPR 2026 · Architectures and efficiency › Pruning and efficient tuning · approach: LoRA variants and theory · 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 LoRA-aware KnOTS-TIES instead of TIESTIES · 8 vision tasks per-tasknormalized accuracy avg (%): 64.9 -> 68.0Improved
use LoRA-LEGO rank-wise clustering instead of vanilla mergingTask Arithmetic · 8 vision tasks per-tasknormalized accuracy avg (%): 63.8 -> 62.9No clear effect
add direction-wise reweighting with shared singular directions (Variant B) over AdaMergingAdaMerging · 8 vision tasks per-tasknormalized accuracy avg (%): 70.0 -> 76.3Improved
TARA Variant B vs AdaMerging on NLIAdaMerging · 6 NLI tasks per-tasknormalized accuracy avg (%): 78.7 -> 80.3Improved
TARA vs AdaMerging on MNLIAdaMerging · 6 NLI tasks per-taskMNLI normalized accuracy: 47.5 -> 46.8Mixed
TARA Variant B vs AdaMerging on joint taskAdaMerging · Joint-task evaluationHits@1: 48.1 -> 49.3Improved
TARA Variant B vs AdaMerging on unseen tasksAdaMerging · Unseen-task generalizationunseen avg accuracy (%): 53.7 -> 57.5Improved
Iso-C style merging vs TARA Variant AIso-C · 8 vision tasks per-tasknormalized accuracy avg (%): 73.6 -> 74.0Improved
TARA preference-weighted merging versus AdaMerging on preference perturbationAdaMerging · GTSRB/SVHN preference samplingtwo-task accuracy spreadImproved

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