知识卡片
架构等变不等于严格不变
卷积的共享局部结构在相应边界与步幅条件下提供平移等变,额外汇总才可能获得一定平移不变性。训练目标也可鼓励指定变换下表示接近。架构约束与经验损失是两种来源,任何一种都不应被笼统说成对所有变换严格不变。
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
- · 表示学习
原书出处 · 1 处
- · 查看原书依据
Learning invariant representations is classic goal of computer vision. Recall that this was one of the reasons we used convolutional image filters: convolution is equivariant with camera translation, and invariance can then be achieved simply by pooling over filter responses. CNNs, through their convolutional architecture, bake translation invariance into the hypothesis space. In this section we will see how to incentivize invariances instead through the
参考来源
- 位置:《Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)》第30章《表示学习》(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/c030b.xhtml)
- 结论依据:摘录中关于架构等变不等于严格不变的定义、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:Learning invariant representations is classic goal of computer vision. Recall that this was one of the reasons we used convolutional image filters: convolution is equivariant with camera translation, and invariance can then be achieved simply by pooling over filter responses. CNNs, through their convolutional architecture, bake translation invariance into the hypothesis space. In this section we will see how to incentivize invariances instead through the