知识卡片
潜在长记忆不是无限存储
循环状态可以经多步传递,使当前输出在计算图上依赖很早的输入。但固定维度、有限精度状态不能无损保存任意长历史,长距离依赖也可能难以学到。理论图灵完备性依赖特定数学模型与资源假设,不能解释为实际任意RNN都能运行所有程序或永不遗忘。
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
- · 循环神经网络
原书出处 · 1 处
- · 查看原书依据
The answer is memory. The recurrent feedback loops in an RNN are a kind of memory. They propagate information from timestep t to timestep t + 1. In other words, they remember information about timestep t when processing new information at timestep t + 1. Thus, by induction, an RNN can potentially remember information over arbitrarily long time intervals. In fact, RNNs are Turing complete:
参考来源
- 位置:《Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)》第25章《循环神经网络》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)/_epub-src/OPS/xhtml/c025.xhtml)
- 结论依据:摘录中关于潜在长记忆不是无限存储的定义、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:The answer is memory. The recurrent feedback loops in an RNN are a kind of memory. They propagate information from timestep t to timestep t + 1. In other words, they remember information about timestep t when processing new information at timestep t + 1. Thus, by induction, an RNN can potentially remember information over arbitrarily long time intervals. In fact, RNNs are Turing complete: