AI Paper Atlas

Research map / Architectures and efficiency / Pruning and efficient tuning

DRONE: Data-aware Low-rank Compression for Large NLP Models

Patrick CHen, Hsiang-Fu Yu, Inderjit S Dhillon, Cho-Jui Hsieh

NeurIPS 2021 · Architectures and efficiency › Pruning and efficient tuning · approach: Joint pruning and quantization · 18 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
replace SVD with data-aware low-rank decomposition at same ranks (ablation)BERT-base · GLUE MRPCF1: 63.8 -> 86.7Improved
replace SVD with data-aware low-rank decomposition at same ranks (ablation)BERT-base · GLUE CoLAMatthew's correlation: 12.4 -> 52.5Improved
add 1-epoch fine-tuning after DRONE (ablation)DRONE · GLUE MRPCF1: 86.7 -> 88.0Improved
add 1-epoch fine-tuning after DRONE (ablation)DRONE · GLUE MNLIaccuracy: 82.0 -> 82.6Improved
use DRONE instead of fine-tuned SVD (ablation)SVD-Retrain · GLUE CoLAMatthew's correlation: 24.4 -> 53.2Improved
combine DRONE with TinyBERT distillationTinyBERT · GLUE SST-2CPU speedup: 11.7x -> 15.3xImproved
combine DRONE with TinyBERT distillationTinyBERT · GLUE STS-Baccuracy: 86.9 -> 87.0No clear effect
DRONE (no attention) vs structured pruningBERT-base · GLUE MRPCaccuracy: 88.61 -> 89.4Improved
use DRONE instead of SVD as low-rank init before trainingRTE training · GLUE RTEaccuracy: 52.1 -> 72.9Improved
DRONE low-rank at same rank as inference setting before trainingRTE training · GLUE RTEaccuracy: 72.6 -> 68.2Hurt
apply DRONE to machine translation transformer2-layer Transformer · en-deBLEU: 33.47 -> 33.26No clear effect

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