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
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 changed | Where | Measured | Effect |
|---|---|---|---|
| use LoRA-aware KnOTS-TIES instead of TIES | TIES · 8 vision tasks per-task | normalized accuracy avg (%): 64.9 -> 68.0 | Improved |
| use LoRA-LEGO rank-wise clustering instead of vanilla merging | Task Arithmetic · 8 vision tasks per-task | normalized accuracy avg (%): 63.8 -> 62.9 | No clear effect |
| add direction-wise reweighting with shared singular directions (Variant B) over AdaMerging | AdaMerging · 8 vision tasks per-task | normalized accuracy avg (%): 70.0 -> 76.3 | Improved |
| TARA Variant B vs AdaMerging on NLI | AdaMerging · 6 NLI tasks per-task | normalized accuracy avg (%): 78.7 -> 80.3 | Improved |
| TARA vs AdaMerging on MNLI | AdaMerging · 6 NLI tasks per-task | MNLI normalized accuracy: 47.5 -> 46.8 | Mixed |
| TARA Variant B vs AdaMerging on joint task | AdaMerging · Joint-task evaluation | Hits@1: 48.1 -> 49.3 | Improved |
| TARA Variant B vs AdaMerging on unseen tasks | AdaMerging · Unseen-task generalization | unseen avg accuracy (%): 53.7 -> 57.5 | Improved |
| Iso-C style merging vs TARA Variant A | Iso-C · 8 vision tasks per-task | normalized accuracy avg (%): 73.6 -> 74.0 | Improved |
| TARA preference-weighted merging versus AdaMerging on preference perturbation | AdaMerging · GTSRB/SVHN preference sampling | two-task accuracy spread | Improved |
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.
- RobustMerge — RobustMerge: Parameter-Efficient Model Merging for MLLMs with Direction Robustness (NeurIPS 2025) · better on 1 of 1 shared result
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
- SDS-LoRA: Overcoming Anisotropic Gradient Scaling in Low-Rank Adaptation (NeurIPS 2026 · similarity 0.90)
- NP-LoRA: Null Space Projection for Subject-Style LoRA Fusion (NeurIPS 2026 · similarity 0.90)
- IterIS: Iterative Inference-Solving Alignment for LoRA Merging (CVPR 2025 · similarity 0.90)
- Crowded in B-Space: Calibrating Shared Directions for LoRA Merging (NeurIPS 2026 · similarity 0.90)
- Accurate and Efficient Low-Rank Model Merging in Core Space (NeurIPS 2025 · similarity 0.89)
- No Task Left Behind: Isotropic Model Merging with Common and Task-Specific Subspaces (ICML 2025 · similarity 0.89)
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.
