← 返回刚才阅读的位置

知识卡片

普通读书笔记卡 · 6486.g.1 · 生成模型:GPT-6

长期记忆依靠门控状态

理解等级:中级

LSTM在循环状态中利用门控调节旧记忆保留、新信息写入与输出,使学习较长依赖更容易。它并非保证任意久的信息都能保留;是否记住与训练、序列长度和门控结果相关。图示提供机制概览,不等于已给出可运行训练方案。

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

      In standard RNNs, this reiterating module will have a straightforward structure, for instance, alone tanh layer. Long Short-Term Memory has a comparative control stream as an intermittent neural organization. The distinctions are the tasks inside the Long Short-Term Memory cells

    参考来源

    - 位置:《Artificial Intelligence and Data Science in Recommendation System: Current Trends, Technologies, and Applications》第07章《电影评论情感分析》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Artificial Intelligence and Data Science in Recommendation System: Current Trends, Technologies, and Applications/_epub-src/OEBPS/Text/9789815136746-C7.html) - 结论依据:摘录中关于长期记忆依靠门控状态的定义、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:In standard RNNs, this reiterating module will have a straightforward structure, for instance, alone tanh layer. Long Short-Term Memory has a comparative control stream as an intermittent neural organization. The distinctions are the tasks inside the Long Short-Term Memory cells