Research map / Architectures and efficiency / Pruning and efficient tuning
Learning in the Fisher Subspace: A Guided Initialization for LoRA Fine-Tuning
Zhi-Quan Feng, Ying-Jia Lin, Hung-Yu Kao
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 low-Fisher-energy directions with Fisher-based scaling instead of SVD singular-value scaling (min directions) (ablation) | FILet · Reasoning avg, Llama2-7B | accuracy (%): 81.9 -> 83.1 | Improved |
| use Fisher-based scaling instead of SVD scaling (ablation) | FILet · Reasoning avg, Llama3-8B | accuracy (%): 85.8 -> 86.7 | Improved |
| select minimal-Fisher-energy directions instead of maximal (ablation) | FILet · Reasoning avg, Llama2-7B | accuracy (%): 81.8 -> 83.1 | Improved |
| select minimal-Fisher-energy directions instead of random (ablation) | FILet · Reasoning avg, Llama3-8B | accuracy (%): 86.4 -> 86.7 | Improved |
| select minimal-Fisher-energy directions instead of random (ablation) | FILet · Image classification avg, ViT-B/32 | accuracy (%): 68.9 -> 69.3 | Improved |
| use minimal (rather than maximal) singular-value scaling in preliminary directions test (ablation) | LoRA init on Llama2-7B · ARC-c and BoolQ | accuracy | Improved |
| increase initialization minibatch size from 80 to Full dataset (ablation) | FILet · Reasoning avg | accuracy (%): up to 0.7% on Llama2-7B and 0.9% on Llama3-8B | Improved |
| increase LoRA rank from 32 to 128 (ablation) | FILet · Reasoning avg, Llama2-7B | accuracy (%): 83.1 -> 82.6 | Hurt |
| increase LoRA rank from 4 to 64 (ablation) | FILet · Reasoning avg, Llama2-7B | accuracy (%): 80.8 -> 83.2 | Improved |
| replace SVD-based or data-driven init (KaSA, LoRA-Dash) with Fisher-guided init | LoRA · GSM8k Llama3-8B | accuracy (%): 55.8 -> 58.8 | Improved |
| replace baselines with Fisher-guided init on ARC-c | LoRA · ARC-c Llama3-8B | accuracy (%): 80.3 -> 79.9 | Hurt |
| Fisher-guided init on image classification vs LoRA | LoRA · ViT-B/32 avg | accuracy (%): 68.4 -> 69.3 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- DoRA — DoRA: Weight-Decomposed Low-Rank Adaptation (ICML 2024) · better on 1 of 1 shared result
- PiSSA — PiSSA: Principal Singular Values and Singular Vectors Adaptation of Large Language Models (NeurIPS 2024) · 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
- Uni-LoRA: One Vector is All You Need (NeurIPS 2025 · similarity 0.91)
- LoRA-One: One-Step Full Gradient Could Suffice for Fine-Tuning Large Language Models, Provably and Efficiently (ICML 2025 · similarity 0.90)
- Loquetier: A Virtualized Multi-LoRA Framework for Unified LLM Fine-tuning and Serving (NeurIPS 2025 · similarity 0.90)
- Beyond LoRA vs. Full Fine-Tuning: Gradient-Guided Optimizer Routing for LLM Adaptation (NeurIPS 2026 · similarity 0.90)
- LoRA vs Full Fine-tuning: An Illusion of Equivalence (NeurIPS 2025 · similarity 0.90)
- SDS-LoRA: Overcoming Anisotropic Gradient Scaling in Low-Rank Adaptation (NeurIPS 2026 · 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.
