Research map / Architectures and efficiency / Pruning and efficient tuning
LoRA-S: An Efficient Low Rank Adaptation scheme via Sylvester equation
Jinyang ZHENG, Tong Wu
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 |
|---|---|---|---|
| lift Adam to horizontal space (AdamS, efficient feature learning) | Adam · Mix-of-Show (r=4) | CLIP score: 25.17 -> 32.20 | Improved |
| lift Adam to horizontal space (AdamS, efficient feature learning) | Adam · Mix-of-Show (r=4) | FID: 72.75 -> 54.39 | Improved |
| use AdamS instead of Scaled GD | Scaled GD · Mix-of-Show (r=16) | CLIP score: 29.07 -> 32.64 | Improved |
| use AdamS instead of Scaled GD | Scaled GD · Mix-of-Show (r=16) | FID: 59.87 -> 46.32 | Improved |
| use AdamS instead of LoRA-Rite | LoRA-Rite · Mix-of-Show (r=4) | CLIP score: 30.96 -> 32.20 | Improved |
| use AdamS instead of Adam | Adam · E2E NLG GPT-2 medium (r=4) | BLEU: 68.0 -> 69.4 | Improved |
| use LRACS instead of AdamW | AdamW · E2E NLG GPT-2 medium (r=4) | BLEU: 68.6 -> 70.4 | Improved |
| use LRACS instead of LoRA-Rite | LoRA-Rite · E2E NLG GPT-2 medium (r=4) | BLEU: 69.3 -> 70.4 | Improved |
| use AdamS instead of Adam | Adam · E2E NLG GPT-2 small (r=4) | BLEU: 68.8 -> 69.4 | Improved |
| add weight-decay regularization of the Sylvester matrix K (ablation) | AdamS · Mix-of-Show (r=16) | CLIP score: 32.64 -> 31.97 | Hurt |
| increase rank from 4 to 16 (ablation) | AdamS · Mix-of-Show | FID: 54.39 -> 46.32 | Improved |
| vary learning rate from 50% to 250% of default (ablation) | AdamS · Mix-of-Show (r=4) | CLIP score: 32.11 -> 31.95 | No clear effect |
Closest papers by meaning
- StelLA: Subspace Learning in Low-rank Adaptation using Stiefel Manifold (NeurIPS 2025 · similarity 0.91)
- LoRA+: Efficient Low Rank Adaptation of Large Models (ICML 2024 · similarity 0.91)
- SDS-LoRA: Overcoming Anisotropic Gradient Scaling in Low-Rank Adaptation (NeurIPS 2026 · similarity 0.90)
- Riemannian Preconditioned LoRA for Fine-Tuning Foundation Models (ICML 2024 · similarity 0.90)
- LoRA-GA: Low-Rank Adaptation with Gradient Approximation (NeurIPS 2024 · similarity 0.90)
- Flat-LoRA: Low-Rank Adaptation over a Flat Loss Landscape (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.
