知识卡片
时间展开共享参数而非独立复制
把递推按多个时刻展开便于构造训练计算图,各时刻仍使用同一组W、U等参数。推理可以逐步在线更新状态,不要求所有未来或全部时刻数据同时到齐;原文对实现的绝对表述需限定为某种批处理训练方式。
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
- · 人工神经网络
原书出处 · 1 处
- · 查看原书依据
This relationship is shown in figure 10.13. As an RNN cell is reusing its internal state ht in the next iteration, a network that looks back N time-steps is modeled as N cells that are laterally connected. As this is how an RNN is actually implemented, the data from all time-steps are presented at the same time.
参考来源
- 位置:《Introduction to Autonomous Robots Mechanisms, Sensors, Actuators, and Algorithms (Nikolaus Correll, Bradley Hayes etc.)》第10章《人工神经网络》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Introduction to Autonomous Robots Mechanisms, Sensors, Actuators, and Algorithms (Nikolaus Correll, Bradley Hayes etc.)/_epub-src/OEBPS/xhtml/chapter_10.xhtml)
- 结论依据:摘录中关于时间展开共享参数而非独立复制的定义、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:This relationship is shown in figure 10.13. As an RNN cell is reusing its internal state ht in the next iteration, a network that looks back N time-steps is modeled as N cells that are laterally connected. As this is how an RNN is actually implemented, the data from all time-steps are presented at the same time.