知识卡片

普通读书笔记卡 · 9174 · 生成模型:GPT-6

展平只改变布局不学习特征

理解等级:中级

Flatten把空间维与通道维按约定排成向量,元素数不变,本身没有可训练参数。28×28×3共有2352个值,但卷积成20通道后的长度还取决于实际空间尺寸,不能把2352同时当任意20特征图的长度。

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

      The first step in previous neural network models has been to flatten a two-dimensional (2D) input image into a one-dimensional (1D) vector. This has been the precondition to apply a dense layer and has been accomplished during preprocessing. However, convolutional neural networks (CNNs) require multidimensional inputs (e.g., 2D images with multiple color channels). Turning a multidimensional tensor into a vector is known as flattening and

    参考来源

    - 位置:《Introduction to Autonomous Robots Mechanisms, Sensors, Actuators, and Algorithms (Nikolaus Correll, Bradley Hayes etc.)》第10章《人工神经网络》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Introduction to Autonomous Robots Mechanisms, Sensors, Actuators, and Algorithms (Nikolaus Correll, Bradley Hayes etc.)/_epub-src/OEBPS/xhtml/chapter_10.xhtml) - 结论依据:摘录中关于展平只改变布局不学习特征的定义、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:The first step in previous neural network models has been to flatten a two-dimensional (2D) input image into a one-dimensional (1D) vector. This has been the precondition to apply a dense layer and has been accomplished during preprocessing. However, convolutional neural networks (CNNs) require multidimensional inputs (e.g., 2D images with multiple color channels). Turning a multidimensional tensor into a vector is known as flattening and