AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

Continuous Self-Attention Models with Neural ODE Networks

Jing Zhang, Peng Zhang, Baiwen Kong, Junqiu Wei, Xin Jiang

AAAI 2021 · Architectures and efficiency › Attention and transformers · approach: Sparse and structured attention · 15 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 CSAODE instead of MPSANtext classification models · CR testaccuracy: 85.4 -> 87Improved
use CSAODE instead of MPSANtext classification models · MPQA testaccuracy: 90.4 -> 90.8Improved
use CSAODE instead of MPSANtext classification models · SUBJ testaccuracy: 94.6 -> 95.9Improved
use CSAODE instead of MPSANtext classification models · TREC testaccuracy: 94.8 -> 95.4Improved
use CSAODE instead of Star-TransformerSNLI · SNLI testaccuracy: 86.0 -> 86.1Improved
use CSAODE instead of CNN+TransformerWikiQA · WikiQA testMRR: 0.703 -> 0.706Improved
remove ARKNet and use one self-attention block (ablation)CSAARK · TREC testaccuracy: 95.2 -> 93.7Hurt
replace the ODE solver with five residual-connected self-attention blocks (ablation)CSAARK · TREC testaccuracy: 95.2 -> 94.2Hurt
add N-gram convolution to capture local information (ablation)CSAARK · TREC testaccuracy: 92.8 -> 95.2Improved
reduce hidden dimension from 300 to 50 (ablation)CSAARK · TREC testaccuracy: 95.2 -> 91.6Hurt
increase hidden dimension from 200 to 300 (ablation)CSAARK · TREC testaccuracy: 94.3 -> 95.2Improved
increase convolution window size from 1 to 2 (ablation)CSAARK · TREC testaccuracy: 94.1 -> 95.2Improved

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