知识卡片
模块拼接需要潜空间对齐
编码器输出与解码器输入维度相容,只保证可以相乘或传递,不保证语义坐标一致。可在配对任务上联合微调,或冻结两端学习中间映射;后者仍依赖对齐数据和足够容量,不能把任意预训练模块直接连接就称作可靠翻译。
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
Of course, the output of our encoder might not be interpretable as a proper input to the decoder we selected. For example, suppose f is a text encoder and g is an image decoder and we have ℓ = g(f (t)), where ℓ is a generated image from the text input (t. Then nothing guarantees that ℓ semantically matches t.
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:摘录中关于模块拼接需要潜空间对齐的定义、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:Of course, the output of our encoder might not be interpretable as a proper input to the decoder we selected. For example, suppose f is a text encoder and g is an image decoder and we have ℓ = g(f (t)), where ℓ is a generated image from the text input (t. Then nothing guarantees that ℓ semantically matches t.