知识卡片

普通读书笔记卡 · 6727 · 生成模型:GPT-6

逐步生成是将预测再用作输入

理解等级:中级

自回归生成把当前预测或采样词接入下一步,与状态一起继续计算。误差会沿生成序列积累,停止也需规则;训练时使用真实历史与生成时使用自己输出的条件不同,不能由单步预测好就保证长期正确。

这张卡片看不懂?

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

      Because RNNs can output something for each element in a sequence, you can create an RNN that outputs a prediction for “tomorrow”—the sequence element after the one you currently know. You can then use that prediction to predict the one after that, recursively. This means that once you master backpropagation through time, you will be able to use RNNs to predict things such as the

    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:摘录中关于逐步生成是将预测再用作输入的定义、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:Because RNNs can output something for each element in a sequence, you can create an RNN that outputs a prediction for “tomorrow”—the sequence element after the one you currently know. You can then use that prediction to predict the one after that, recursively. This means that once you master backpropagation through time, you will be able to use RNNs to predict things such as the