• LSTM block和cell区别


    LSTM的结构中每个时刻的隐层包含了多个memory blocks(一般我们采用一个block),每个block包含了包含一个Cell(有多个memory cell组成)和三个gate,一个基础的结构示例如下图: 
    image

    一个memory cell只能产出一个标量值,一个block能产出一个向量。

     The nomenclature is a bit confusing, but goes back to the original LSTM paper: an LSTM "cell" is strictly speaking mostly the CEC, i.e., the "inner" unit that stores a value through time. A "block" is CEC + gates. In theory, you could have more than one cell in a block. In that case, all cells within the block would use the same gates. So they would store/output information together. This way you could e.g. store two "bits" of information with the same gates. However this idea never really caught on -- all modern LSTMs have one cell per block, and the distinction between cell & blocks eroded over time (which luckily simplifies notation, which was often a bit cumbersome in old Hochreiter, Gers and Graves papers that used the idea of blocks).

    so, cell应该是指LSTM内部的四个神经网络层(全连接层)中的神经元,而block显然是包含了这些神经网络层和门的看起来是方块的整体。

  • 相关阅读:
    NoSql数据库简介及Redis学习
    C++内存泄露
    实现堆排、快排、归并
    常见C/C++笔试、面试题(二)
    Linux的五种I/O模式
    设计模式之Iterator模式
    MapReduce简介
    PHP字符串函数试题
    PHP之curl函数相关试题
    PHP数学函数试题
  • 原文地址:https://www.cnblogs.com/pomodoro/p/9272608.html
Copyright © 2020-2023  润新知