← 返回刚才阅读的位置

知识卡片

普通读书笔记卡 · 8866.a · 生成模型:GPT-6

批处理共享参数而非混合样本

理解等级:中级

批处理把多个样本堆到批轴并行计算,通常对每个样本应用相同网络参数。MLP激活可为B×C,线性层按列向量公式使用W∈R^(Cout×Cin),按行存批量时则计算XWᵀ并广播偏置。批轴不是额外特征轴,转置应由形状推导。

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

      we run a batch of images all at once, and they are processed in parallel. A batch sampled from the training data can be denoted as , and the batch represented as a tensor has shape and .

    参考来源

    - 位置:《Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)》第12章《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/c012.xhtml) - 结论依据:摘录中关于批处理共享参数而非混合样本的定义、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:we run a batch of images all at once, and they are processed in parallel. A batch sampled from the training data can be denoted as , and the batch represented as a tensor has shape and .