Research map / Label-efficient and robust learning / Semi-supervised learning
Revisiting LSTM Networks for Semi-Supervised Text Classification via Mixed Objective Function
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 |
|---|---|---|---|
| train with the mixed objective instead of maximum likelihood alone | one-layer BiLSTM · ACL-IMDB | error rate (%): 6.43 -> 4.32 | Helped |
| train with the mixed objective instead of maximum likelihood alone | one-layer BiLSTM · AG-News | error rate (%): 5.62 -> 4.95 | Helped |
| add adversarial training to ML using labeled data only (ablation) | BiLSTM on ACL-IMDB · ACL-IMDB | error rate (%): 6.43 -> 5.96 | Helped |
| add entropy minimization to ML using labeled data only (ablation) | BiLSTM on ACL-IMDB · ACL-IMDB | error rate (%): 6.43 -> 6.46 | No clear effect |
| add virtual adversarial training to ML using labeled data only (ablation) | BiLSTM on ACL-IMDB · ACL-IMDB | error rate (%): 6.43 -> 5.98 | Helped |
| add unlabeled data to the EM and VAT objectives (ablation) | BiLSTM on ACL-IMDB · ACL-IMDB | error rate (%): 5.52 -> 4.47 | Helped |
| finetune pretrained embeddings rather than keep them static (ablation) | BiLSTM trained with ML · ACL-IMDB | error rate (%): 8.17 -> 6.43 | Helped |
| use a larger vocabulary instead of limiting it to 30,000 words (ablation) | BiLSTM trained with ML · ACL-IMDB | error rate (%): 7.78 -> 6.43 | Helped |
| remove text preprocessing (ablation) | BiLSTM trained with ML · ACL-IMDB | error rate (%): 6.43 -> 8.80 | Hurt |
| increase BiLSTM depth to 3 layers (ablation) | L_MIXED · ACL-IMDB | error rate (%): 4.32 -> 4.15% | Helped |
| increase BiLSTM depth to 4 layers (ablation) | L_MIXED · ACL-IMDB | error rate | Hurt |
| use an ensemble of separately trained objective models rather than joint training | BiLSTM on ACL-IMDB · ACL-IMDB | error rate (%): 4.3% vs 5.2% | Hurt |
Closest papers by meaning
- Supervised and Semi-Supervised Text Categorization using LSTM for Region Embeddings (ICML 2016 · similarity 0.87)
- Automatic Rule Extraction from Long Short Term Memory Networks (ICLR 2017 · similarity 0.86)
- Mixture of Expert/Imitator Networks: Scalable Semi-Supervised Learning Framework (AAAI 2019 · similarity 0.85)
- Adversarial Training Methods for Semi-Supervised Text Classification (ICLR 2017 · similarity 0.85)
- Dual Adversarial Co-Learning for Multi-Domain Text Classification (AAAI 2020 · similarity 0.84)
- TACIT: A Target-Agnostic Feature Disentanglement Framework for Cross-Domain Text Classification (AAAI 2024 · similarity 0.82)
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.