知识卡片

普通读书笔记卡 · 8974 · 生成模型:GPT-6

共现视图可以使用不同编码器

理解等级:中级

音频与图像、灰度与色彩等同源信号可分别编码,并把同一样本的两种视图对齐,与不匹配视图区分。不同模态输入维度可以不同,输出嵌入维度需适合匹配。共现通常携带共同信息,但并不保证共享的恰是所需语义。

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

      A second kind of contrastive learning is based on co-occurrence, where the goal is to learn a common representation of all co-occurring signals. This form of contrastive learning is useful for learning, for example, an audiovisual representation where the embedding of an image matches the embedding of the sound for that same scene. Or, returning to our colorization example, we can learn an image representation

    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:摘录中关于共现视图可以使用不同编码器的定义、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:A second kind of contrastive learning is based on co-occurrence, where the goal is to learn a common representation of all co-occurring signals. This form of contrastive learning is useful for learning, for example, an audiovisual representation where the embedding of an image matches the embedding of the sound for that same scene. Or, returning to our colorization example, we can learn an image representation