AI Paper Atlas

Research map / Continual learning / Domain and prompt-based

Gradient Regularized Contrastive Learning for Continual Domain Adaptation

Shixiang Tang, Peng Su, Dapeng Chen, Wanli Ouyang

AAAI 2021 · Continual learning › Domain and prompt-based · approach: Domain-incremental learning with memory · 46 citations (OpenAlex)

This paper tried a common change: Add a contrastive loss

Add contrastive learning to source-only supervised training · Digits, DomainNet, and Office-Caltech target domains · classification accuracy Improved

What 1,317 accepted papers report for this change

See this change in detailMatched by the wording of each paper's own result, read by AI from the full text. Papers report gains more often than failures, so this shows what gets published, not the chance a change works.

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
Replace ADDA-based alignment with contrastive learning while retaining gradient constraintsGRA · DigitsACC: 84.10 -> 85.34Improved
Replace ADDA-based alignment with contrastive learning while retaining gradient constraintsGRA · DomainNetACC: 35.84 -> 37.74Improved
Replace ADDA-based alignment with contrastive learning while retaining gradient constraintsGRA · Office-CaltechACC: 86.53 -> 87.23Improved
Increase memory size per domain from 256 to 2048 (ablation)GRCL · DigitsACC: 83.00 -> 85.41Improved
Increase memory size per domain from 256 to 2048 (ablation)GRCL · DomainNetACC: 33.28 -> 37.83Improved
Increase training schedule from 120 to 240 epochs (ablation)GRCL · DigitsACC: 80.10 -> 85.34Improved
Increase training schedule from 120 to 240 epochs (ablation)GRCL · DomainNetACC: 34.80 -> 37.74Improved
Increase training schedule from 120 to 240 epochs (ablation)GRCL · Office-CaltechACC: 80.93 -> 87.23Improved
Train beyond 240 epochs, increasing the schedule to 300 (ablation)GRCL · DigitsACC: 85.34 -> 85.38No clear effect
Train beyond 240 epochs, increasing the schedule to 300 (ablation)GRCL · Office-CaltechACC: 87.23 -> 87.28No clear effect
Add source-loss gradient regularization to contrastive learning (ablation)Multi-task contrastive and supervised learning · Digits, DomainNet, and Office-Caltech target domainsclassification accuracyImproved
Add contrastive learning to source-only supervised training (ablation)Source-only model · Digits, DomainNet, and Office-Caltech target domainsclassification accuracyImproved

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