Research map / Image generation / Autoregressive and latent generation
Beyond Adapter Retrieval: Latent Geometry-Preserving Composition via Sparse Task Projection
Pengfei Jin, Peng Shu, Sifan Song, Sekeun Kim, Qing Xiao, Cheng Chen, Tianming Liu, Xiang Li, Quanzheng Li
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 l1 sparse projection instead of uniform averaging | adapter composition · Prostate-E zero-shot | Dice: 64.6% -> 79.1% | Improved |
| use l1 sparse projection instead of uniform averaging | adapter composition · AbdAtlas zero-shot | Dice: 19.2% -> 64.5% | Improved |
| add l1 sparsity to least-squares combination (ablation) | adapter composition · Prostate-E zero-shot | Dice: 52.0% -> 79.1% | Improved |
| add l1 sparsity to least-squares combination (ablation) | adapter composition · AbdAtlas zero-shot | Dice: 2.5% -> 64.5% | Improved |
| compose multiple adapters instead of nearest only (ablation) | adapter composition · Prostate-E | Dice: 0.1% -> 79.1% | Improved |
| compose multiple adapters instead of nearest only (ablation) | adapter composition · Prostate-A | Dice: 90.5% -> 90.5% | No clear effect |
| use sparse projection vs least-squares on report generation | adapter composition · CT Abdomen ROUGE-L | ROUGE-L: 0.1393 -> 0.1379 | Hurt |
| use sparse projection vs uniform average on report generation | adapter composition · X-ray ROUGE-L | ROUGE-L: 0.1776 -> 0.2084 | Improved |
| add in-domain supervised LoRA to candidate pool (ablation) | sparse composition · Prostate-C | Dice: 90.5% -> 90.8% | Improved |
| use sparse projection for SD v1.5 adapter composition | adapter composition · MS COCO 100 prompts (SD v1.5) | preference win rate | Improved |
Closest papers by meaning
- C-LoRA: Continual Low-Rank Adaptation for Pre-trained Visual Models (NeurIPS 2026 · similarity 0.84)
- Latent Space Factorization in LoRA (NeurIPS 2025 · similarity 0.84)
- StelLA: Subspace Learning in Low-rank Adaptation using Stiefel Manifold (NeurIPS 2025 · similarity 0.84)
- LoRAtorio: An intrinsic approach to LoRA Skill Composition (NeurIPS 2026 · similarity 0.84)
- Variational Task Vector Composition (NeurIPS 2025 · similarity 0.84)
- When the Merge Coefficient Stops Mattering: Proximity Regularized Merging for Continual LoRA Adaptation (NeurIPS 2026 · similarity 0.84)
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.
