Group-wise Contrastive Learning for Neural Dialogue Generation 閱讀筆記
摘要 爲解決生成多樣性低的問題 思想 找到正向和負向group的context和response，設計loss 做法 首先有 其中pn是baseline模型，pm是target模型 其中c是輸入文本 r是回覆文本， 構造loss： 最大化正向回覆，最小化負向回覆 這個loss即刻畫 相對baseline模型的生成多樣性 或 分佈distinctiveness 正向回覆和負向回覆的採樣，使用了BM2
相關文章
- 1. Deep Reinforcement Learning for Dialogue Generation閱讀筆記
- 2. Adversarial Learning for Neural Dialogue Generation 代碼分析
- 3. Deep Reinforcement Learning for Dialogue Generation
- 4. Deep Reinforcement Learning for Dialogue Generation 論文閱讀 A Diversity-Promoting Objective Function fo
- 5. 論文分享-- >Adversarial Learning for Neural Dialogue Generation
- 6. 論文閱讀筆記《Meta-Learning of Neural Architectures for Few-Shot Learning》
- 7. 閱讀《Learning to Ask: Neural Question Generation for Reading Comprehension 》
- 8. 論文閱讀筆記《Attentive Weights Generation for Few Shot Learning via Information Maximization》
- 9. 《Contextualized Code Representation Learning for Commit Message Generation》閱讀
- 10. Matching Networks for one Shot Learning 閱讀筆記
- 更多相關文章...
- • RSS 閱讀器 - RSS 教程
- • PHP 實例 - AJAX RSS 閱讀器 - PHP教程
- • Tomcat學習筆記（史上最全tomcat學習筆記）
- • JDK13 GA發佈：5大特性解讀
相關標籤/搜索
每日一句
-
每一个你不满意的现在，都有一个你没有努力的曾经。
歡迎關注本站公眾號,獲取更多信息
相關文章
- 1. Deep Reinforcement Learning for Dialogue Generation閱讀筆記
- 2. Adversarial Learning for Neural Dialogue Generation 代碼分析
- 3. Deep Reinforcement Learning for Dialogue Generation
- 4. Deep Reinforcement Learning for Dialogue Generation 論文閱讀 A Diversity-Promoting Objective Function fo
- 5. 論文分享-- >Adversarial Learning for Neural Dialogue Generation
- 6. 論文閱讀筆記《Meta-Learning of Neural Architectures for Few-Shot Learning》
- 7. 閱讀《Learning to Ask: Neural Question Generation for Reading Comprehension 》
- 8. 論文閱讀筆記《Attentive Weights Generation for Few Shot Learning via Information Maximization》
- 9. 《Contextualized Code Representation Learning for Commit Message Generation》閱讀
- 10. Matching Networks for one Shot Learning 閱讀筆記