知识卡片
共享参数梯度需要累加
若每步都有监督损失,总成本是各步损失之和,同一参数在不同时间位置出现的梯度贡献也要相加。某步状态既影响当前输出又影响未来状态,反传在该分支处同样累加。不能只取最后一次调用的参数梯度或覆盖早先贡献。
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
What about the gradient of the total cost with respect to the parameters, ? Suppose the RNN is being applied to a video and predicts a class label for every frame of the video. Then xout is a prediction for each frame and a loss can be applied to each xout[t]. The total cost is simply the summation of the losses at each timestep:
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:摘录中关于共享参数梯度需要累加的定义、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:What about the gradient of the total cost with respect to the parameters, ? Suppose the RNN is being applied to a video and predicts a class label for every frame of the video. Then xout is a prediction for each frame and a loss can be applied to each xout[t]. The total cost is simply the summation of the losses at each timestep: