Research map / Label-efficient and robust learning / Recognition and long-tail
Contrastive Learning with Adversarial Perturbations for Conditional Text Generation
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 CLAPS adversarial positive and negative examples | T5-small · WMT'16 RO-EN | BLEU: 32.43 -> 33.96 | Helped |
| train with CLAPS adversarial positive and negative examples | T5-small · XSum test | ROUGE-1: 36.10 -> 37.89 | Helped |
| train with CLAPS adversarial positive and negative examples | T5-small · SQuAD test | BLEU: 21.00 -> 21.55 | Helped |
| add adversarial negative examples (ablation) | T5-CLAPS · WMT'16 RO-EN | BLEU: 33.50 -> 33.96 | Helped |
| add distant-target positive examples (ablation) | T5-CLAPS · WMT'16 RO-EN | BLEU: 32.55 -> 33.96 | Helped |
| add adversarial negative examples (ablation) | T5-CLAPS · XSum test | ROUGE-1: 37.49 -> 37.89 | Helped |
| add distant-target positive examples (ablation) | T5-CLAPS · XSum test | ROUGE-1: 37.72 -> 37.89 | Helped |
| add adversarial negative examples (ablation) | T5-CLAPS · SQuAD test | BLEU: 21.38 -> 21.55 | Helped |
| add distant-target positive examples (ablation) | T5-CLAPS · SQuAD test | BLEU: 21.25 -> 21.55 | Helped |
| use random batch-shuffled negative pairs for contrastive training | T5-MLE-contrastive · WMT'16 RO-EN | BLEU: 32.43 -> 32.03 | Hurt |
| use random batch-shuffled negative pairs for contrastive training | T5-MLE-contrastive · SQuAD test | BLEU: 21.00 -> 20.91 | Hurt |
| perturb target representations to make imposters, rather than perturbing them toward the source embedding | CLAPS · conditional text generation | performance | Helped |
Closest papers by meaning
- MaskGAN: Better Text Generation via Filling in the _______ (ICLR 2018 · similarity 0.87)
- Seq2Sick: Evaluating the Robustness of Sequence-to-Sequence Models with Adversarial Examples (AAAI 2020 · similarity 0.87)
- SCOPE: A Self-supervised Framework for Improving Faithfulness in Conditional Text Generation (ICLR 2025 · similarity 0.85)
- Error Norm Truncation: Robust Training in the Presence of Data Noise for Text Generation Models (ICLR 2024 · similarity 0.85)
- Customizing Language Model Responses with Contrastive In-Context Learning (AAAI 2024 · similarity 0.84)
- Tractable Transformers for Flexible Conditional Generation (ICML 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.