AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

Multi-Scale Self-Attention for Text Classification

Qipeng Guo, Xipeng Qiu, Pengfei Liu, Xiangyang Xue, Zheng Zhang

AAAI 2020 · Architectures and efficiency › Attention and transformers · approach: Sparse and structured attention · 76 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 standard self-attention with multi-scale self-attentionTransformer · SST testaccuracy: 50.4 -> 51.9Improved
Replace standard self-attention with multi-scale self-attentionTransformer · MTL-16 averageaccuracy: 82.78 -> 86.34Improved
Replace standard self-attention with multi-scale self-attentionTransformer · PTB POS, CoNLL2003 NER, CoNLL2012 NERaccuracy / F1: 96.31 -> 97.24; 86.48 -> 89.38; 83.57 -> 85.26Improved
Replace standard self-attention with multi-scale self-attentionTransformer · SNLI testaccuracy: 82.2 -> 85.9Improved
Set the scale-distribution factor α to 0.5, favoring smaller scales in shallow layers (ablation)Multi-Scale Transformer · SNLI testaccuracy: 84.9 -> 85.9Improved
Set α to -0.5 or -1.0, increasing global bias in shallow layers (ablation)Multi-Scale Transformer · SNLI testaccuracy: 85.9 -> 84.7; 85.9 -> 84.3Hurt
Use a single fixed attention scale instead of multiple scales (ablation)Multi-Scale Transformer · SNLI testaccuracy: 85.9 -> 84.3 (ω=3); 85.9 -> 83.9 (ω=N/16); 85.9 -> 81.7 (ω=N/8); 85.9 -> 80.7 (ω=N/4)Hurt
Add character-level features (ablation)Multi-Scale Transformer · PTB POS / CoNLL2003 NER / CoNLL2012 NERaccuracy / F1: 97.24 -> 97.54; 89.38 -> 91.33; 85.26 -> 86.77Improved
Add a CRF layer (ablation)Multi-Scale Transformer + Char · PTB POS / CoNLL2003 NER / CoNLL2012 NERaccuracy / F1: 97.54 -> 97.66; 91.33 -> 91.59; 86.77 -> 87.80Improved
Add the classification node and combine it with max-poolingMulti-Scale Transformer · Text classification and SNLIaccuracyImproved

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