AI Paper Atlas

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

Simpson’s Bias in NLP Training

Fei Yuan, Longtu Zhang, Huang Bojun, Yaobo Liang

AAAI 2021 · Label-efficient and robust learning › Recognition and long-tail · approach: Group-robust classification · 0 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
Train BERT with Dice loss instead of CE lossBERT-base-uncased · MRPC devF1: 89.78 -> 89.00Hurt
Train BERT with Dice loss instead of CE lossBERT-base-uncased · QQP devF1: 87.84 -> 87.39Hurt
Train BERT-base multilingual cased with Dice loss instead of CE lossBERT-base multilingual cased · GermEval 2014 testF1: 86.14 -> 82.61Hurt
Train with the averaged sample-level objective rather than the corpus-level metricDice-loss training · MRPCF1: + 0.78Hurt
Train with the averaged sample-level objective rather than the corpus-level metricDice-loss training · QQPF1: + 0.45Hurt
Train with the averaged sample-level objective rather than the corpus-level metricDice-loss training · NERF1: 3.53 pointHurt
Use the alternative token-independent Dice loss formulation (ablation)NER Dice-loss training · GermEval 2014Simpson's biasHurt
Use smoothing constant gamma=1 (ablation)Dice-loss training · PSM and NERSimpson's biasNo clear effect
Increase Simpson's bias significanceDice-loss training · MRPC trainF1Hurt
Change learning rate, batch size, or model type (ablation)Dice-loss training on MRPC · MRPCSimpson's biasNo clear effect

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