AI Paper Atlas

Research map / Architectures and efficiency / Pruning and efficient tuning

Sparse Structure Search for Delta Tuning

Shengding Hu, Zhen Zhang, Ning Ding, Yadao Wang, Yasheng Wang, Zhiyuan Liu, Maosong Sun

NeurIPS 2022 · Architectures and efficiency › Pruning and efficient tuning · approach: Adapter-based parameter-efficient tuning · 0 citations (OpenAlex)

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
Search sparse structure with S3Delta-M (1.39% params) instead of manual BitFit (8.13% params)T5-large delta tuning · GLUEaverage score: 86.90 -> 87.17Improved
Search with S3Delta-M instead of manual LoRA (r=1) at fewer parametersT5-large delta tuning · GLUEaverage score: 86.84 -> 87.17Improved
Search with S3Delta-M instead of manual BitFitT5-large delta tuning · SuperGLUE excluding COPAaverage score: 82.41 -> 83.10Improved
Search with S3Delta-M instead of manual LoRA (r=1)T5-large delta tuning · SuperGLUE excluding COPAaverage score: 81.88 -> 83.10Improved
Narrow the search space from Mix to LoRA only (S3Delta-L vs S3Delta-M) (ablation)S3Delta · GLUEaverage score: 87.17 -> 86.91Hurt
Narrow the search space from Mix to LoRA only (S3Delta-L vs S3Delta-M) (ablation)S3Delta · SuperGLUEaverage score: 81.30 -> 79.26Hurt
Search LoRA positions (S3Delta-L) instead of uniform LoRA (r=1)LoRA · SuperGLUE excluding COPAaverage score: 81.88 -> 82.46Improved
Reduce trainable parameters of S3Delta-M from 1.39% to 0.35% (ablation)S3Delta-M · GLUEaverage score: 87.17 -> 86.43Hurt
Reduce trainable parameters of S3Delta-M from 1.39% to 0.35% (ablation)S3Delta-M · SuperGLUE excluding COPAaverage score: 83.10 -> 81.24Hurt
Use S3Delta-M structure instead of random structures at extremely low budgets (0.086%)S3Delta-M · RTE, MultiRC, MRPCFT performance recovered: 96.8%,98.7%,97.5%Improved
Increase trainable budget to 5.6%all methods · MultiRC, MRPCscoreNo clear effect
Replace L0 regularization with shifted global sigmoid for sparsity control (ablation)S3Delta · MRPC, RTE, MultiRCscoreImproved

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