AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

Learning Language Representations with Logical Inductive Bias

Jianshu Chen

ICLR 2023 · Architectures and efficiency › Attention and transformers · approach: Efficient language transformers · 0 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 FOLNet Base instead of BERT Baselanguage representation pretraining · GLUE dev, 16GB pretrainingAvg: 83.1 -> 87.7Improved
use FOLNet Base instead of DeBERTa Baselanguage representation pretraining · SQuAD 2.0 dev, 16GB pretrainingEM: 79.3 -> 84.7Improved
use FOLNet Base instead of DeBERTa Baselanguage representation pretraining · SQuAD 2.0 dev, 16GB pretrainingF1: 82.5 -> 87.9Improved
use FOLNet Base instead of RoBERTa Baselanguage representation pretraining · GLUE dev, 160GB pretrainingAvg: 86.4 -> 89.2Improved
use FOLNet Base instead of RoBERTa Baselanguage representation pretraining · FOLIO validation, 160GB pretrainingaccuracy: 64.7 -> 64.2Hurt
use FOLNet Large instead of RoBERTa Largelanguage representation pretraining · FOLIO validation, 160GB pretrainingaccuracy: 67.7 -> 70.6Improved
use the dual-branch j.a architecture instead of BERT (ablation)FOLNet · GLUE devAvg: 83.1 -> 84.0Improved
use proposed relative positional encoding instead of absolute positional encoding (ablation)FOLNet j.a · GLUE devAvg: 84.0 -> 84.7Improved
replace proposed relative positional encoding with T5 relative positional encoding (ablation)FOLNet j.a · GLUE devAvg: 84.7 -> 84.7No clear effect
add mu and prod operators (ablation)FOLNet j.a · GLUE devAvg: 84.7 -> 85.4Improved
increase binary predicate dimension D2 from 12 to 64 (ablation)FOLNet j.a · GLUE devAvg: 84.0 -> 85.1Improved
add trans, prod, mu, and c operators to j.a (ablation)FOLNet · GLUE devAvg: 85.1 -> 87.7Improved

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