AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

Towards Efficient Pre-Trained Language Model via Feature Correlation Distillation

Kun Huang, Xin Guo, Meng Wang

NeurIPS 2023 · Architectures and efficiency › Attention and transformers · approach: Language model pretraining and induction · 1 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 Pearson linear correlation instead of KL divergence (ablation)FCD · MNLI/QQP/RTEaverage: 75.1 -> 75.8Improved
use Pearson linear correlation instead of MSE (ablation)FCD · MNLI/QQP/RTEaverage: 75.5 -> 75.8Improved
use uniform layer selection instead of top layers (ablation)FCD · MNLI-maccuracy: 82.9 -> 83.8Improved
use uniform layer selection instead of top layers (ablation)FCD · CoLAMatthews corr.: 50.7 -> 52.0Improved
use uniform layer selection instead of bottom layers (ablation)FCD · MNLI-maccuracy: 83.3 -> 83.8Improved
use uniform layer selection instead of bottom layers (ablation)FCD · CoLAMatthews corr.: 51.5 -> 52.0Improved
remove token-level relation loss (ablation)FCD · SST-2scoreHurt
remove sample-level relation loss (ablation)FCD · GLUE (QNLI, MNLI-m, MRPC, SST-2)scoreHurt
distill with FCD instead of KD from BERT-base to 6 layers (ablation)BERT-small6 · GLUE test averageAVG score: 74.8 -> 79.6Improved
use FCD instead of MiniLMv2 (BERT teacher)6-layer student · GLUE test averageAVG score: 79.2 -> 79.6Improved
use FCD instead of MiniLMv2 (RoBERTa teacher)6-layer student · GLUE test averageAVG score: 81.0 -> 81.9Improved
use FCD instead of original KD on decoder modelDistilGPT2 · GLUE test averageAVG score: 75.3 -> 77.1Improved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

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