知识卡片
双向编码只适合未来输入可用场景
双向RNN结合正向与反向上下文,适合整句已知的编码或标注。在线逐词预测未来时不能偷看之后真词,否则泄漏;Transformer也不因此就是带bidirectional开关的循环模型,效果提升无统一5%到10%保证。
这张卡片看不懂?
这张卡片已通过人工审查,不再接受失败反馈。
已有反馈 · 0 条
还没有人提交反馈。
在整书笔记中的位置
原书出处 · 1 处
- · 查看原书依据
Of course, the PyTorch implementation has many other features. PyTorch RNNs can even be trained from left to right and right to left simultaneously! Of course, for a bidirectional language model like this to be of any use, your problem must be noncausal. In English-language NLP, a noncausal model is simply a type of model designed to predict words that occur before (to the left
参考来源
- 位置:《Natural Language Processing in Action, Second Edition》第08章《循环网络与长短期记忆》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Natural Language Processing in Action, Second Edition/_epub-src/OEBPS/Text/chapter-8.html)
- 结论依据:摘录中关于双向编码只适合未来输入可用场景的定义、比较或条件支撑卡片;原书表述不严谨的范围已在覆盖记录逐项限定,不把案例结果外推为普遍结论。
- 原始内容:Of course, the PyTorch implementation has many other features. PyTorch RNNs can even be trained from left to right and right to left simultaneously! Of course, for a bidirectional language model like this to be of any use, your problem must be noncausal. In English-language NLP, a noncausal model is simply a type of model designed to predict words that occur before (to the left