AI Paper Atlas

Research map / Architectures and efficiency / CNN design

ACT: an Attentive Convolutional Transformer for Efficient Text Classification

Pengfei Li, Peixiang Zhong, Kezhi Mao, Dongzhe Wang, Xuefeng Yang, Yunfeng Liu, Jianxiong Yin, Simon See

AAAI 2021 · Architectures and efficiency › CNN design · approach: Filter design and trained-filter analysis · 39 citations (OpenAlex)

Read the paper

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 attentive convolution instead of conventional CNN feature maps (ablation)ACT · Yelp F. developmentaccuracy (%): 67.1 -> 68.3Improved
Use attentive convolution instead of conventional CNN feature maps (ablation)ACT · TACRED developmentmicro-averaged F1: 66.5 -> 67.8Improved
Add the multi-head structure (ablation)ACT · Yelp F. developmentaccuracy (%): 67.0 -> 68.3Improved
Add the multi-head structure (ablation)ACT · TACRED developmentmicro-averaged F1: 65.9 -> 67.8Improved
Include global representation in global attention (ablation)ACT · Yelp F. developmentaccuracy (%): 67.6 -> 68.3Improved
Include global representation in global attention (ablation)ACT · TACRED developmentmicro-averaged F1: 67.1 -> 67.8Improved
Include position embeddings in global attention (ablation)ACT · Yelp F. developmentaccuracy (%): 67.4 -> 68.3Improved
Include position embeddings in global attention (ablation)ACT · TACRED developmentmicro-averaged F1: 63.5 -> 67.8Improved
Use ACT instead of Transformertext classification models · Yelp F. testaccuracy (%): 65.34* -> 68.16Improved
Use ACT instead of Transformertext classification models · AGNews testaccuracy (%): 93.89* -> 94.25Improved
Use ACT instead of Transformertext classification models · DBPedia testaccuracy (%): 98.98* -> 99.19Improved
Use ACT instead of Transformertext classification models · Yelp P. testaccuracy (%): 96.13* -> 97.41Improved

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