Research map / Architectures and efficiency / Pruning and efficient tuning
S2FT: Parameter-Efficient Fine-Tuning in Sparse Spectrum Domain
Baoquan Zhang, Zhehao Yu, Lisai Zhang, Kenghong Lin, Tianran Chen, Yuxi Sun, Yunming Ye, Yao He
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 S2FT instead of FourierFT (half the parameters) | ViT-B/16 · VTAB-1k | mean accuracy: 72.8 -> 73.6 | Improved |
| use S2FT instead of FourierFT (same 0.16% parameters) | ViT-B/16 · VTAB-1k | mean accuracy: 72.8 -> 74.1 | Improved |
| use S2FT instead of FourierFT | ViT-B/16 · FGVC | mean accuracy: 87.9 -> 89.1 | Improved |
| use S2FT instead of FourierFT | RoBERTa-base · GLUE | average score: 85.0 -> 85.6 | Improved |
| use S2FT instead of FourierFT | RoBERTa-large · GLUE | average score: 88.0 -> 88.3 | Improved |
| use S2FT instead of FourierFT | LLaMA2-13B · MT-Bench | GPT-4 score: 5.82 -> 5.89 | Improved |
| use S2FT instead of FourierFT | LLaMA2-7B · MT-Bench | GPT-4 score: 5.18 -> 5.21 | No clear effect |
| use spectrum-prior sampling instead of random sampling (ablation) | S2FT · VTAB-1k | mean accuracy: 72.0 -> 73.6 | Improved |
| use spectrum-prior sampling instead of low-frequency sampling (ablation) | S2FT · VTAB-1k | mean accuracy: 72.8 -> 73.6 | Improved |
| increase gamma from 0.5 to 1.5 (ablation) | S2FT · VTAB-1k | mean accuracy: 72.3 -> 73.6 | Improved |
| increase gamma from 1.5 to 5.0 (ablation) | S2FT · VTAB-1k | mean accuracy: 73.6 -> 72.7 | Hurt |
| pre-estimate weight change with full fine-tuning (5 epochs) instead of cumulative gradient (ablation) | S2FT · VTAB-1k | mean accuracy: 73.6 -> 73.8 | No clear effect |
Later papers that compared with this method
1 later accepted paper reports S2FT as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- Breaking Information Islands in Sparse Tuning via Small-World Connectivity (NeurIPS 2026 · better on 1 of 1 shared result)
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.
- 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
- Spectral Adapter: Fine-Tuning in Spectral Space (NeurIPS 2024 · similarity 0.87)
- S$^{2}$FT: Efficient, Scalable and Generalizable LLM Fine-tuning by Structured Sparsity (NeurIPS 2024 · similarity 0.87)
- Parameter-Efficient Fine-Tuning with Discrete Fourier Transform (ICML 2024 · similarity 0.87)
- Expanding Sparse Tuning for Low Memory Usage (NeurIPS 2024 · similarity 0.87)
- F-Adapter: Frequency-Adaptive Parameter-Efficient Fine-Tuning in Scientific Machine Learning (NeurIPS 2025 · similarity 0.86)
- SVFT: Parameter-Efficient Fine-Tuning with Singular Vectors (NeurIPS 2024 · similarity 0.85)
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.
