Research map / Label-efficient and robust learning / Semi-supervised learning
Variational Autoencoder for Semi-Supervised Text Classification
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 |
|---|---|---|---|
| Feed the class label to the decoder at every time step rather than only as the initial input (ablation) | SSVAE decoder · IMDB | test error rate: 17.76% -> 10.28% | Helped |
| Feed the class label to the decoder at every time step rather than only as the initial input (ablation) | SSVAE decoder · AG's News | test error rate: 12.69% -> 9.71% | Helped |
| Use CLSTM-II, which directly adds label information to the memory cell, instead of CLSTM-I (ablation) | SSVAE · IMDB and AG's News | test error rate: 10.38% -> 10.28%; 10.22% -> 9.71% | Helped |
| Initialize the classifier with LM-LSTM pretraining | SSVAE-II · IMDB | test error rate: 8.72% -> 7.23% | Helped |
| Initialize the classifier with LM-LSTM pretraining | SSVAE-II · AG's News | test error rate: 9.71% -> 8.97%; 9.12% -> 8.33%; 8.30% -> 7.60% | Helped |
| Estimate the unlabeled gradient by sampling with the sequence-length-dependent S1 baseline (ablation) | SSVAE-II · IMDB | test error rate: 10.28% -> 16.87% | Hurt |
| Estimate the unlabeled gradient by sampling with the sequence-length-dependent S1 baseline (ablation) | SSVAE-II · AG's News | test error rate: 9.71% -> 11.92% | Hurt |
| Use the S2 baseline, averaging sampled label objectives, for sampled gradient estimation (ablation) | SSVAE-II · AG's News | test error rate: 9.71% -> 9.89% | No clear effect |
| Replace full class enumeration with S1 sampling (ablation) | SSVAE-II training · AG's News, 1 epoch | training time: 1880(s) -> 1070(s) | Helped |
| Replace full class enumeration with S2 sampling (ablation) | SSVAE-II training · AG's News, 1 epoch | training time: 1880(s) -> 1205(s) | Helped |
Closest papers by meaning
- Infinite Variational Autoencoder for Semi-Supervised Learning (CVPR 2017 · similarity 0.87)
- Unified Robust Semi-Supervised Variational Autoencoder (ICML 2021 · similarity 0.84)
- Variational Autoencoder for Deep Learning of Images, Labels and Captions (NeurIPS 2016 · similarity 0.84)
- Rethinking Controllable Variational Autoencoders (CVPR 2022 · similarity 0.83)
- Consistency Regularization for Variational Auto-Encoders (NeurIPS 2021 · similarity 0.83)
- Enhancing Semi-Supervised Learning via Representative and Diverse Sample Selection (NeurIPS 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.