Research map / Label-efficient and robust learning / Domain adaptation and generalization
Improving Domain-Adapted Sentiment Classification by Deep Adversarial Mutual Learning
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 changed | Where | Measured | Effect |
|---|---|---|---|
| Add label probers that pass peer-classifier target predictions to each group's feature extractor | DAML · Average performance | Acc: 63.4% to 64.5% | Helped |
| Add label probers that pass peer-classifier target predictions to each group's feature extractor | DAML · Average performance | RMSE: 0.925 to 0.898 | Helped |
| Train probers on source domain only rather than target domain only (ablation) | DAML · Yelp → IMDB | Acc: 0.563 -> 0.521 | Hurt |
| Train probers on both source and target domains rather than target domain only (ablation) | DAML · Yelp → IMDB | Acc: 0.563 -> 0.546 | Hurt |
| Add target-domain prober training rather than training probers on source domain only (ablation) | DAML · Yelp → IMDB | Acc: 0.521 -> 0.563 | Helped |
| Use standard mutual learning, directly aligning peer classifiers with target predictions, instead of label probers (ablation) | DAML · Domain-adaptation tasks | Acc | Hurt |
| Align the two models' feature representations with Euclidean distance (ablation) | DAML · Domain-adaptation tasks | Acc | Hurt |
| Use a naive ensemble of two domain-adapted models instead of the DAML single-model evaluation (ablation) | DAML · Average performance | Acc: 63.4% to 64.5% | Hurt |
| Increase the number of DAML groups from 2 to 3 (ablation) | DAML · Yelp → IMDB | Acc: 0.563 -> 0.563 | No clear effect |
| Increase the number of DAML groups from 2 to 4 (ablation) | DAML · Yelp → IMDB | Acc: 0.563 -> 0.547 | Hurt |
| Increase the number of DAML groups from 2 to 3 (ablation) | DAML · IMDB → Yelp | Acc: 0.528 -> 0.531 | Helped |
| Increase the number of DAML groups from 2 to 4 (ablation) | DAML · IMDB → Yelp | Acc: 0.528 -> 0.522 | Hurt |
Closest papers by meaning
- Adversarial Training Based Multi-Source Unsupervised Domain Adaptation for Sentiment Analysis (AAAI 2020 · similarity 0.89)
- Multi-Source Domain Adaptation for Visual Sentiment Classification (AAAI 2020 · similarity 0.89)
- Dual Adversarial Co-Learning for Multi-Domain Text Classification (AAAI 2020 · similarity 0.88)
- An Adaptive Hybrid Framework for Cross-domain Aspect-based Sentiment Analysis (AAAI 2021 · similarity 0.88)
- Adversarial Multiple Source Domain Adaptation (NeurIPS 2018 · similarity 0.86)
- Simultaneous Learning of Pivots and Representations for Cross-Domain Sentiment Classification (AAAI 2020 · similarity 0.85)
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.