知识卡片

普通读书笔记卡 · 8855.a · 生成模型:GPT-6

动量累积历史更新方向

理解等级:中级

动量可写vₖ₊₁=μvₖ−η∇J(θₖ),再令θₖ₊₁=θₖ+vₖ₊₁。持续同向的梯度会累积速度,短期方向变化的影响被平滑。μ与学习率共同决定行为,不能只看当前梯度推断下一步位移。

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

      Momentum makes the analogy to skiing even more precise: momentum is like the inertia of the skier, carrying them over the little bumps and imperfections in the ski slope and increasing their speed as they descend along a straight path. In math, momentum just means that we set the parameter update to be a direction vk+1, given by a weighted combination of the previous update

    参考来源

    - 位置:《Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)》第10章《基于梯度的学习算法》(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/c010.xhtml) - 结论依据:摘录中关于动量累积历史更新方向的定义、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:Momentum makes the analogy to skiing even more precise: momentum is like the inertia of the skier, carrying them over the little bumps and imperfections in the ski slope and increasing their speed as they descend along a straight path. In math, momentum just means that we set the parameter update to be a direction vk+1, given by a weighted combination of the previous update