Papers with bottom-k sampling
PCC: Paraphrasing with Bottom-k Sampling and Cyclic Learning for Curriculum Data Augmentation (2023.eacl-main)
Copied to clipboard
| Challenge: | Curriculum Data Augmentation (CDA) presents synthetic data with increasing difficulties to neural models. |
| Approach: | They propose a curriculum-aware paraphrase generation module with bottom-k sampling and cyclic learning strategy that passes through the curriculums multiple times. |
| Outcome: | The proposed framework surpasses competitive baselines on few-shot text classification and dialogue generation. |