Research map / Optimization / Optimizers
LoRA-DA: Data-Aware Initialization for Low-Rank Adaptation via Asymptotic Analysis
Qingyue Zhang, Chang Chu, Tianren Peng, Qi Li, Xiangyang Luo, Zhihao Jiang, Shao-Lun Huang
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-DA data-aware init instead of MiLoRA | LoRA (LLaMA 2-7B) · Commonsense reasoning avg | accuracy: 84.0 -> 84.3 | Improved |
| use LoRA-DA init instead of LoRA-One | LoRA (LLaMA 2-7B) · GSM8K | exact match: 53.7 -> 55.0 | Improved |
| use LoRA-DA init instead of LoRA-One | LoRA (LLaMA 2-7B) · MATH | exact match: 8.5 -> 9.2 | Improved |
| use LoRA-DA init instead of LoRA-One | LoRA (LLaMA 2-7B) · Math avg | average accuracy: 31.1 -> 32.1 | Improved |
| use LoRA-DA init instead of PiSSA | LoRA (LLaMA 2-13B) · GSM8K | exact match: 65.8 -> 66.4 | Improved |
| use LoRA-DA init instead of MiLoRA | LoRA (LLaMA 2-13B) · MATH | exact match: 12.2 -> 12.8 | Improved |
| use LoRA-DA init instead of random LoRA-FA init | LoRA-FA (frozen A) · GSM8K | exact match: 41.5 -> 49.4 | Improved |
| use LoRA-DA init instead of PiSSA | LoRA (LLaMA 2-7B) · Commonsense reasoning avg | accuracy: 83.5 -> 84.3 | Improved |
| add the bias term to LoRA-DA initialization (ablation) | LoRA-DA · GSM8K | accuracy | Improved |
| add the variance term to LoRA-DA initialization (ablation) | LoRA-DA · GSM8K | accuracy | Improved |
| use Fisher-gradient instead of plain gradient in the bias term (ablation) | LoRA-DA · GSM8K | accuracy | Improved |
| increase LoRA rank from small to large (ablation) | LoRA-DA vs LoRA and LoRA-One · GSM8K | accuracy | Mixed |
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.
- LoRA-One — LoRA-One: One-Step Full Gradient Could Suffice for Fine-Tuning Large Language Models, Provably and Efficiently (ICML 2025) · better on 6 of 6 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
- LoRA-GA: Low-Rank Adaptation with Gradient Approximation (NeurIPS 2024 · similarity 0.94)
- The Impact of Initialization on LoRA Finetuning Dynamics (NeurIPS 2024 · similarity 0.91)
- LoRA Training Provably Converges to a Low-Rank Global Minimum Or It Fails Loudly (But it Probably Won't Fail) (ICML 2025 · similarity 0.90)
- Beyond Zero Initialization: Investigating the Impact of Non-Zero Initialization on LoRA Fine-Tuning Dynamics (ICML 2025 · similarity 0.90)
- SDS-LoRA: Overcoming Anisotropic Gradient Scaling in Low-Rank Adaptation (NeurIPS 2026 · similarity 0.90)
- LoRA-Pro: Are Low-Rank Adapters Properly Optimized? (ICLR 2025 · similarity 0.90)
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.
