知识卡片

普通读书笔记卡 · 6738.b · 生成模型:GPT-6

门控单元不按固定倍数衡量容量

理解等级:中级

GRU和LSTM有更多门与参数,提供不同状态更新路径,但不等于每个单元恰好两个普通RNN神经元。隐藏宽度相同也不保证同容量或质量;应统计实际参数、耗时及验证表现,再讨论效率。

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

      When you are thinking about how many units to add to your neural network to solve a particular problem, each LSTM or GRU unit gives your network a capacity similar to two “normal” RNN neurons or hidden vector dimensions. A unit is just a more complicated, higher-capacity neuron, which you can see by counting up the number of learned parameters in your LSTM model and

    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:摘录中关于门控单元不按固定倍数衡量容量的定义、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:When you are thinking about how many units to add to your neural network to solve a particular problem, each LSTM or GRU unit gives your network a capacity similar to two “normal” RNN neurons or hidden vector dimensions. A unit is just a more complicated, higher-capacity neuron, which you can see by counting up the number of learned parameters in your LSTM model and