知识卡片
上下文学习改变状态而非权重
RNN 依次处理 (x1,y1,…,xn,yn) 后形成隐藏状态,再回答新查询 x_(n+1)。在这次任务中权重可保持固定,行为仍由历史样例改变;训练这个“根据样例学习”的序列模型可视为元学习。状态适配与参数梯度更新应分开描述。
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
- · 迁移学习与适配
原书出处 · 1 处
- · 查看原书依据
RNNs update their hidden states as they process a sequence of data. These changes in hidden state change the behavior of the mapping from input to output. Therefore, RNNs adapt as they process data. Now consider that learning is the problem of adapting future behavior as a function of past data. RNNs can be run on sequences of frames in a movie or on sequences
参考来源
- 位置:《Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)》第37章《迁移学习与适配》(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/c037.xhtml)
- 结论依据:摘录中关于上下文学习改变状态而非权重的定义、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:RNNs update their hidden states as they process a sequence of data. These changes in hidden state change the behavior of the mapping from input to output. Therefore, RNNs adapt as they process data. Now consider that learning is the problem of adapting future behavior as a function of past data. RNNs can be run on sequences of frames in a movie or on sequences