Research map / Architectures and efficiency / Pruning and efficient tuning
Towards a Unified View of Parameter-Efficient Transfer Learning
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 parallel rather than sequential adapter insertion at the FFN (ablation) | Adapter · XSum | ROUGE-2: 19.98 -> 20.66 | Helped |
| Use parallel rather than sequential adapter insertion at the FFN (ablation) | Adapter · WMT16 en-ro | BLEU: 36.3 -> 37.1 | Helped |
| Modify FFN rather than attention (ablation) | LoRA · XSum | ROUGE-2: LoRA (ffn) improves LoRA (attn) by 1 R-2 points | Helped |
| Use multi-head parallel adapters rather than single-head parallel adapters (ablation) | Parallel Adapter · WMT16 en-ro | BLEU: +1.6 BLEU points | Helped |
| Retain gating in prefix tuning rather than replacing it with simple addition (ablation) | Prefix tuning · WMT16 en-ro | BLEU: +0.3 points | Helped |
| Reduce prefix bottleneck dimension from 200 to 30 (ablation) | Prefix tuning · WMT16 en-ro | BLEU: 0.4 BLEU loss | Hurt |
| Add scaled composition to parallel adapter (ablation) | Parallel Adapter · XSum | ROUGE-2: 20.98 -> 21.54 | Helped |
| Use a trainable scalar instead of fixed scaling (ablation) | Scaled Parallel Adapter · XSum | ROUGE-2: 21.54 -> 21.31 | Hurt |
| Combine small-prefix attention tuning with scaled FFN adapters (ablation) | Parameter-efficient tuning · XSum | ROUGE-2: 21.71 -> 21.90 | Helped |
| Combine small-prefix attention tuning with scaled FFN adapters (ablation) | Parameter-efficient tuning · WMT16 en-ro | BLEU: 37.0 -> 37.5 | Helped |
| Use parameter-efficient tuning instead of full fine-tuning | MAM Adapter · MNLI | accuracy: 87.6 -> 87.4 | No clear effect |
| Use parameter-efficient tuning instead of full fine-tuning | MAM Adapter · SST2 | accuracy: 94.6 -> 94.2 | No clear effect |
Closest papers by meaning
- Seeking Neural Nuggets: Knowledge Transfer in Large Language Models from a Parametric Perspective (ICLR 2024 · similarity 0.88)
- Enhancing Large Language Model Performance with Gradient-Based Parameter Selection (AAAI 2025 · similarity 0.88)
- When Scaling Meets LLM Finetuning: The Effect of Data, Model and Finetuning Method (ICLR 2024 · similarity 0.87)
- Data Mixing Optimization for Supervised Fine-Tuning of Large Language Models (ICML 2025 · similarity 0.87)
- Evaluating the Zero-shot Robustness of Instruction-tuned Language Models (ICLR 2024 · similarity 0.87)
- A Paradigm Shift in Machine Translation: Boosting Translation Performance of Large Language Models (ICLR 2024 · 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.