AI Paper Atlas

Research map / Architectures and efficiency / Pruning and efficient tuning

A Win-win Deal: Towards Sparse and Robust Pre-trained Language Models

Yuanxin Liu, Fandong Meng, Zheng Lin, Jiangnan Li, Peng Fu, Yanan Cao, Weiping Wang, Jie Zhou

NeurIPS 2022 · Architectures and efficiency › Pruning and efficient tuning · approach: Low-rank compression and merging · 2 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 hard mask initialization with soft (ablation)mask training at 90% sparsity · QQPaccuracy: 71.64 -> 77.08Improved
replace hard mask initialization with soft (ablation)mask training at 90% sparsity · FEVERaccuracy: 49.56 -> 72.80Improved
replace hard mask initialization with soft (ablation)mask training at 90% sparsity · MNLIaccuracy: 72.09 -> 72.63No clear effect
gradual sparsity increase 0.7~0.9 instead of fixed 90% (soft) (ablation)mask training · MNLIaccuracy: 72.63 -> 76.84Improved
gradual sparsity increase 0.7~0.9 instead of fixed 90% (soft) (ablation)mask training · HANSaccuracy: 52.82 -> 56.72Improved
gradual sparsity increase 0.7~0.9 instead of fixed 90% (soft) (ablation)mask training · QQPaccuracy: 77.08 -> 79.49Improved
gradual sparsity increase 0.7~0.9 instead of fixed 90% (soft) (ablation)mask training · FEVERaccuracy: 72.80 -> 79.01Improved
gradual sparsity increase 0.7~0.9 instead of fixed 90% (soft) (ablation)mask training · PAWS-qqpaccuracy: 46.48 -> 46.59No clear effect
start gradual sparsity at 0.7 instead of 0.2 (ablation)mask training at 90% sparsity · MNLIaccuracy: 73.61 -> 76.84Improved
gradual sparsity 50% to 70% instead of fixed 70%mask training PoE · RoBERTa-base MNLIaccuracy: 83.48 -> 84.41Improved
gradual sparsity 50% to 70% instead of fixed 70%mask training PoE · BERT-large MNLIaccuracy: 77.54 -> 84.83Improved
gradual sparsity 50% to 70% instead of fixed 70%mask training PoE · BERT-large HANSaccuracy: 60.19 -> 70.18Improved

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