AI Paper Atlas

Research map / Architectures and efficiency / Pruning and efficient tuning

Compacter: Efficient Low-Rank Hypercomplex Adapter Layers

Rabeeh Karimi mahabadi, James Henderson, Sebastian Ruder

NeurIPS 2021 · Architectures and efficiency › Pruning and efficient tuning · approach: Adapter-based parameter-efficient tuning · 82 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
Use shared hypercomplex weights and layer-specific rank-one factors in adaptersT5-Base fine-tuning on GLUE · GLUEaverage score: 86.50 -> 86.62Improved
Use shared hypercomplex weights and layer-specific rank-one factors in adaptersT5-Base fine-tuning on SuperGLUE · SuperGLUEaverage score: 70.06 -> 71.57Improved
Remove the Compacter layer after self-attention (ablation)Compacter on GLUE · GLUEaverage score: 86.62 -> 86.47No clear effect
Remove the Compacter layer after self-attention (ablation)Compacter on SuperGLUE · SuperGLUEaverage score: 71.57 -> 71.82Improved
Set n=8 instead of n=4 (ablation)Compacter on GLUE · GLUEaverage score: 86.62 -> 86.34Hurt
Set n=12 instead of n=4 (ablation)Compacter on GLUE · GLUEaverage score: 86.62 -> 86.17Hurt
Set n=8 instead of n=4 (ablation)Compacter++ on GLUE · GLUEaverage score: 86.47 -> 85.95Hurt
Set n=12 instead of n=4 (ablation)Compacter++ on GLUE · GLUEaverage score: 86.47 -> 86.34Hurt
Set n=8 instead of n=12 (ablation)Compacter on SuperGLUE · SuperGLUEaverage score: 71.57 -> 71.19Hurt
Set n=4 instead of n=12 (ablation)Compacter on SuperGLUE · SuperGLUEaverage score: 71.57 -> 70.18Hurt
Set n=8 instead of n=12 (ablation)Compacter++ on SuperGLUE · SuperGLUEaverage score: 71.82 -> 68.69Hurt
Set n=4 instead of n=12 (ablation)Compacter++ on SuperGLUE · SuperGLUEaverage score: 71.82 -> 70.53Hurt

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