Research map / Architectures and efficiency / Pruning and efficient tuning
FRoD: Full-Rank Efficient Fine-Tuning with Rotational Degrees for Fast Convergence
Guoan Wan, Tianyu Chen, Fangzheng Feng, Haoyi Zhou, Runhua Xu
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 |
|---|---|---|---|
| add off-axis sparse rotation S to on-axis Sigma updates (ablation) | FRoD · 4 vision datasets, epoch 10 | average accuracy: 71.26 -> 74.59 | Improved |
| add on-axis Sigma updates to off-axis S updates (ablation) | FRoD · 4 vision datasets, epoch 10 | average accuracy: 73.12 -> 74.59 | Improved |
| increase on-axis lr(Sigma) to 5e-03 without off-axis update (ablation) | FRoD on-axis only · 4 vision datasets | average accuracy: epoch 1: 25.89 -> 50.03; epoch 10: 70.77 -> 67.28 | Mixed |
| increase off-axis learning rate lr(S) from 1e-04 to 5e-04 (ablation) | FRoD off-axis only · 4 vision datasets, epoch 10 | average accuracy: 73.12 -> 60.60 | Hurt |
| increase sparsity s from 0.02 to 0.1 (lr(S)=1e-04) (ablation) | FRoD off-axis only · 4 vision datasets, epoch 10 | average accuracy: 73.12 -> 72.77 | No clear effect |
| use FRoD instead of LoRA | CLIP ViT-B/32 · Image classification average (7 datasets) | accuracy: 77.09 -> 82.80 | Improved |
| use FRoD instead of VeRA | LLaMA2-7B · Commonsense reasoning average (6 datasets) | accuracy: 66.70 -> 83.53 | Improved |
| use FRoD instead of full fine-tuning | LLaMA2-7B · Commonsense reasoning average (6 datasets) | accuracy: 84.58 -> 83.53 | Hurt |
| use FRoD instead of full fine-tuning | RoBERTa-large · GLUE average (7 tasks) | score: 90.42 -> 90.53 | No clear effect |
| keep tan(alpha) of off-axis to on-axis rotation in moderate range (ablation) | FRoD · 4 vision datasets, epoch 10 | accuracy | 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.
- VeRA — VeRA: Vector-based Random Matrix Adaptation (ICLR 2024) · better on 2 of 2 shared results
- DoRA — DoRA: Weight-Decomposed Low-Rank Adaptation (ICML 2024) · better on 2 of 2 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
- Parameter Efficient Fine-tuning via Explained Variance Adaptation (NeurIPS 2025 · similarity 0.89)
- AuroRA: Breaking Low-Rank Bottleneck of LoRA with Nonlinear Mapping (NeurIPS 2025 · similarity 0.89)
- Beyond LoRA vs. Full Fine-Tuning: Gradient-Guided Optimizer Routing for LLM Adaptation (NeurIPS 2026 · similarity 0.88)
- GraLoRA: Granular Low-Rank Adaptation for Parameter-Efficient Fine-Tuning (NeurIPS 2025 · similarity 0.88)
- Flat-LoRA: Low-Rank Adaptation over a Flat Loss Landscape (ICML 2025 · similarity 0.88)
- Make LoRA Great Again: Boosting LoRA with Adaptive Singular Values and Mixture-of-Experts Optimization Alignment (ICML 2025 · similarity 0.87)
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.
