AI Paper Atlas

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

ICML 2026 · Optimization › Optimizers · approach: Natural gradient and Fisher approximations

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-DA data-aware init instead of MiLoRALoRA (LLaMA 2-7B) · Commonsense reasoning avgaccuracy: 84.0 -> 84.3Improved
use LoRA-DA init instead of LoRA-OneLoRA (LLaMA 2-7B) · GSM8Kexact match: 53.7 -> 55.0Improved
use LoRA-DA init instead of LoRA-OneLoRA (LLaMA 2-7B) · MATHexact match: 8.5 -> 9.2Improved
use LoRA-DA init instead of LoRA-OneLoRA (LLaMA 2-7B) · Math avgaverage accuracy: 31.1 -> 32.1Improved
use LoRA-DA init instead of PiSSALoRA (LLaMA 2-13B) · GSM8Kexact match: 65.8 -> 66.4Improved
use LoRA-DA init instead of MiLoRALoRA (LLaMA 2-13B) · MATHexact match: 12.2 -> 12.8Improved
use LoRA-DA init instead of random LoRA-FA initLoRA-FA (frozen A) · GSM8Kexact match: 41.5 -> 49.4Improved
use LoRA-DA init instead of PiSSALoRA (LLaMA 2-7B) · Commonsense reasoning avgaccuracy: 83.5 -> 84.3Improved
add the bias term to LoRA-DA initialization (ablation)LoRA-DA · GSM8KaccuracyImproved
add the variance term to LoRA-DA initialization (ablation)LoRA-DA · GSM8KaccuracyImproved
use Fisher-gradient instead of plain gradient in the bias term (ablation)LoRA-DA · GSM8KaccuracyImproved
increase LoRA rank from small to large (ablation)LoRA-DA vs LoRA and LoRA-One · GSM8KaccuracyMixed

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