Research map / Image generation / Autoregressive and latent generation
Dual Supervised 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 probabilistic duality regularization by jointly training both translation directions | RNNSearch · WMT14 En↔Fr | BLEU: 29.92 -> 31.99; 27.49 -> 28.35 | Helped |
| Add probabilistic duality regularization by jointly training both translation directions | RNNSearch · WMT14 En↔De | BLEU: 16.54 -> 17.91; 20.69 -> 20.81 | Helped |
| Add probabilistic duality regularization by jointly training both translation directions | RNNSearch · NIST2008 and NIST2012 En↔Zh | BLEU: 15.45 -> 15.87; 31.67 -> 33.59; 15.05 -> 16.10; 30.54 -> 32.00 | Helped |
| Add probabilistic duality regularization | NMT with UNK post-processing · En→Fr | BLEU: 34.23 -> 34.84 | Helped |
| Add probabilistic duality regularization with conditional image generation | ResNet-32 · CIFAR-10 image classification | error rate (%): 7.51 -> 6.82 | Helped |
| Add probabilistic duality regularization with conditional image generation | ResNet-110 · CIFAR-10 image classification | error rate (%): 6.43 -> 5.40 | Helped |
| Use ResNet-110 classification as the dual task during training | PixelCNN++ · CIFAR-10 image generation | bits per dimension: 2.94 -> 2.93 | Helped |
| Use ResNet-32 classification as the dual task during training | PixelCNN++ · CIFAR-10 image generation | bits per dimension | No clear effect |
| Jointly train sentiment classification and sentiment-conditioned sentence generation | LSTM classifier · IMDB | test error (%): 10.10 -> 9.20 | Helped |
| Jointly train sentiment classification and sentiment-conditioned sentence generation | LSTM sentence generator · IMDB | perplexity: 59.19 -> 58.78 | Helped |
| Use a warm-start generator and randomly initialized classifier (ablation) | DSL · IMDB sentiment classification | test error (%): 10.10 -> 9.44% | Helped |
| Use a warm-start classifier and randomly initialized generator (ablation) | DSL · IMDB sentence generation | perplexity: 59.19 -> 58.79 | Helped |
Closest papers by meaning
- Model-Level Dual Learning (ICML 2018 · similarity 0.90)
- Dual Learning for Machine Translation (NeurIPS 2016 · similarity 0.87)
- Multi-Agent Dual Learning (ICLR 2019 · similarity 0.86)
- Dual Transfer Learning for Neural Machine Translation with Marginal Distribution Regularization (AAAI 2018 · similarity 0.86)
- TANDEM: Bi-Level Data Mixture Optimization with Twin Networks (NeurIPS 2025 · similarity 0.83)
- DualOptim: Enhancing Efficacy and Stability in Machine Unlearning with Dual Optimizers (NeurIPS 2025 · similarity 0.83)
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.