知识卡片
单位化使相似度成为余弦
L2单位向量点积等于cosθ,不是与角度θ成正比。更大点积对应更小夹角,等价于更小单位向量欧氏距离。归一化限制向量范数的缩放捷径,温度或logit缩放仍影响softmax分布与训练梯度。
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
- · 视觉与语言
原书出处 · 1 处
- · 查看原书依据
The dot product between unit vectors is proportional to the angle between the vectors. Figure 51.3 shows how a trained CLIP model maps data to embeddings. This figure is generated in the same way as figure 13.9: for each encoder we reduce dimensionality for visualization using t-distributed Stochastic Neighbor Embedding (t-SNE) [312]. For the image encoder, we show the visual content using icons to reflect
参考来源
- 位置:《Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)》第51章《视觉与语言》(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/c051.xhtml)
- 结论依据:摘录中关于单位化使相似度成为余弦的定义、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 dot product between unit vectors is proportional to the angle between the vectors. Figure 51.3 shows how a trained CLIP model maps data to embeddings. This figure is generated in the same way as figure 13.9: for each encoder we reduce dimensionality for visualization using t-distributed Stochastic Neighbor Embedding (t-SNE) [312]. For the image encoder, we show the visual content using icons to reflect