AI Paper Atlas

Research map / Label-efficient and robust learning / Recognition and long-tail

Merging Statistical Feature via Adaptive Gate for Improved Text Classification

Xianming Li, Zongxi Li, Haoran Xie, Qing Li

AAAI 2021 · Label-efficient and robust learning › Recognition and long-tail · approach: Multilingual language representation pretraining · 41 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
add corpus-level statistical features using the adaptive gateBert-base Uncased · Subjaccuracy: 97.03 -> 97.89Improved
add corpus-level statistical features using the adaptive gateBert-base Uncased · SST-1accuracy: 53.78 -> 55.72Improved
add corpus-level statistical features using the adaptive gateBert-base Uncased · SST-2accuracy: 91.07 -> 93.27Improved
add corpus-level statistical features using the adaptive gateBert-base Uncased · TRECaccuracy: 96.48 -> 97.65Improved
add corpus-level statistical features using the adaptive gateBert-base Uncased · AG’s Newsaccuracy: 93.30 -> 93.82Improved
add corpus-level statistical features using the adaptive gateBert-base Uncased · Yelp F.accuracy: 95.73 -> 96.79Improved
add corpus-level statistical features using the adaptive gateBert-base Uncased · Yelp P.accuracy: 68.46 -> 70.74Improved
use a VAE rather than a vanilla autoencoder to encode TCoL (ablation)CNN+AGN · SST-2, TREC, Yelp F., and Yelp P.accuracy: 81.98 -> 82.24; 92.02 -> 92.11; 94.89 -> 95.15; 67.11 -> 67.93Improved
adaptively filter corrupted statistical information rather than accept all of it (ϵ=0.2 vs ϵ=0.5) (ablation)CNN+AGN · SST-2 and TREC with shuffled TCoLaccuracy: 78.25 -> 80.05; 87.69 -> 89.64Improved
adaptively use only some statistical information instead of rejecting all or accepting all (ablation)AGN · SST-2 and TRECaccuracyImproved

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