{"version":2,"relationshipURLBase":"https://codingai-lec03.pages.dev/course/blocks/retrieval-approximation-and-interaction.html","relationshipResolution":"Resolve relative relationship URLs against relationshipURLBase; complete records and absolute URLs are in blocks.","title":"选读：省搜索和增加图文交互，是不同方向","url":"https://codingai-lec03.pages.dev/course/blocks/retrieval-approximation-and-interaction.html","markdown":"https://codingai-lec03.pages.dev/course/blocks/retrieval-approximation-and-interaction.md","pages":[{"id":"reading-opening","title":"选读：省搜索和增加图文交互，是不同方向","unit":"检索到底复用了什么，又省去了什么？","unitId":"retrieval-cache-and-evaluation","chapter":"III","number":1}],"aliases":[{"oldId":"b-06ae2b95bccfea8a2eeea2d4","targetId":"b-06ae2b95bccfea8a2eeea2d4","page":"reading-opening","operation":"move","beforeSHA256":"07d6d6277d7ef396b98a3e84502c5cfa47a9cd9d0d57844acc5f76a5daac5f97","afterSHA256":"07d6d6277d7ef396b98a3e84502c5cfa47a9cd9d0d57844acc5f76a5daac5f97","reason":"Same source body in authored semantic page"},{"oldId":"b-9a306ce13fe5423c3fcd3b6d","targetId":"b-9a306ce13fe5423c3fcd3b6d","page":"reading-opening","operation":"move","beforeSHA256":"513b4dde07c19aa85a702ba5af2c592cb8b134338c12e49e6c6f6faf9adf92ed","afterSHA256":"513b4dde07c19aa85a702ba5af2c592cb8b134338c12e49e6c6f6faf9adf92ed","reason":"Same source body in authored semantic page"},{"oldId":"b-9cd8158ff12624e30dd37e7c","targetId":"b-9cd8158ff12624e30dd37e7c","page":"reading-opening","operation":"move","beforeSHA256":"f726d4dfd12721dd615d742d7afe75d70e549a32421c9fb296bd5365d8942054","afterSHA256":"f726d4dfd12721dd615d742d7afe75d70e549a32421c9fb296bd5365d8942054","reason":"Same source body in authored semantic page"},{"oldId":"b-ea6996e655977b7b3e7aae39","targetId":"b-ea6996e655977b7b3e7aae39","page":"reading-opening","operation":"move","beforeSHA256":"1766e359ec4d31dd1478ef54c245a79bf9e3cd2bcb748e45f6e24d363cc4b064","afterSHA256":"1766e359ec4d31dd1478ef54c245a79bf9e3cd2bcb748e45f6e24d363cc4b064","reason":"Same source body in authored semantic page"},{"oldId":"b-29527c419f693494a5470d00","targetId":"b-29527c419f693494a5470d00","page":"reading-opening","operation":"move","beforeSHA256":"dfb749be9ea401298591c13840469fdf0010697e73c244978111ebf419a560a3","afterSHA256":"dfb749be9ea401298591c13840469fdf0010697e73c244978111ebf419a560a3","reason":"Same source body in authored semantic page"},{"oldId":"b-7177cf80bc859c52ca80b6b2","targetId":"b-7177cf80bc859c52ca80b6b2","page":"reading-opening","operation":"move","beforeSHA256":"6424f8550f5b5c6628f1965c5c1b4aafa939764425130e58f1a53d555fcca12a","afterSHA256":"6424f8550f5b5c6628f1965c5c1b4aafa939764425130e58f1a53d555fcca12a","reason":"Same source body in authored semantic page"},{"oldId":"b-686882a361bdee07a6ced45b","targetId":"b-686882a361bdee07a6ced45b","page":"reading-opening","operation":"move","beforeSHA256":"4317d93511e6997230f0f90631867ff02fa98c7c130893b0bfe4a4e45f7238d1","afterSHA256":"4317d93511e6997230f0f90631867ff02fa98c7c130893b0bfe4a4e45f7238d1","reason":"Same source body in authored semantic page"},{"oldId":"b-76b28b9ffc1b109f32ac949e","targetId":"b-76b28b9ffc1b109f32ac949e","page":"reading-opening","operation":"move","beforeSHA256":"2b9ccef7e1220e2250cceff1abd8bdd736b42f8da5a223f26b029f44b962e807","afterSHA256":"2b9ccef7e1220e2250cceff1abd8bdd736b42f8da5a223f26b029f44b962e807","reason":"Same source body in authored semantic page"}],"blocks":[{"id":"b-06ae2b95bccfea8a2eeea2d4","type":"h1","scope":"retrieval-approximation-and-interaction","url":"https://codingai-lec03.pages.dev/course/blocks/retrieval-approximation-and-interaction.html#b-06ae2b95bccfea8a2eeea2d4","content":"# 选读：省搜索和增加图文交互，是不同方向","sha256":"07d6d6277d7ef396b98a3e84502c5cfa47a9cd9d0d57844acc5f76a5daac5f97","chapter":{"id":"III","title":"视觉与语言"},"unit":{"id":"retrieval-cache-and-evaluation","title":"检索到底复用了什么，又省去了什么？"},"page":{"id":"reading-opening","title":"选读：省搜索和增加图文交互，是不同方向","number":1},"sectionTitle":"选读：省搜索和增加图文交互，是不同方向","currentQuestion":"选读：省搜索和增加图文交互，是不同方向","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":null,"nextBlock":"b-9a306ce13fe5423c3fcd3b6d","prerequisites":[{"id":"b-9a306ce13fe5423c3fcd3b6d","url":"#b-9a306ce13fe5423c3fcd3b6d","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-9a306ce13fe5423c3fcd3b6d","type":"p","scope":"retrieval-approximation-and-interaction","url":"https://codingai-lec03.pages.dev/course/blocks/retrieval-approximation-and-interaction.html#b-9a306ce13fe5423c3fcd3b6d","content":"后续课程预备或按需选读，不列为本讲必讲内容。","sha256":"513b4dde07c19aa85a702ba5af2c592cb8b134338c12e49e6c6f6faf9adf92ed","chapter":{"id":"III","title":"视觉与语言"},"unit":{"id":"retrieval-cache-and-evaluation","title":"检索到底复用了什么，又省去了什么？"},"page":{"id":"reading-opening","title":"选读：省搜索和增加图文交互，是不同方向","number":1},"sectionTitle":"选读：省搜索和增加图文交互，是不同方向","currentQuestion":"选读：省搜索和增加图文交互，是不同方向","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-06ae2b95bccfea8a2eeea2d4","nextBlock":"b-9cd8158ff12624e30dd37e7c","prerequisites":[{"id":"b-06ae2b95bccfea8a2eeea2d4","url":"#b-06ae2b95bccfea8a2eeea2d4","role":"section-title"}],"relatedBlocks":[]},{"id":"b-9cd8158ff12624e30dd37e7c","type":"p","scope":"retrieval-approximation-and-interaction","url":"https://codingai-lec03.pages.dev/course/blocks/retrieval-approximation-and-interaction.html#b-9cd8158ff12624e30dd37e7c","content":"检索把同一个比较空间用于一组已有图片。图库向量可复用，查询只需编码一次再搜索；页面前三张结果来自已有真实缓存。工作量、相关性指标和近似索引在说明与 LAB12 中展开；检索系统的进一步组织留待第八讲。","sha256":"f726d4dfd12721dd615d742d7afe75d70e549a32421c9fb296bd5365d8942054","chapter":{"id":"III","title":"视觉与语言"},"unit":{"id":"retrieval-cache-and-evaluation","title":"检索到底复用了什么，又省去了什么？"},"page":{"id":"reading-opening","title":"选读：省搜索和增加图文交互，是不同方向","number":1},"sectionTitle":"选读：省搜索和增加图文交互，是不同方向","currentQuestion":"选读：省搜索和增加图文交互，是不同方向","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-9a306ce13fe5423c3fcd3b6d","nextBlock":"b-ea6996e655977b7b3e7aae39","prerequisites":[{"id":"b-06ae2b95bccfea8a2eeea2d4","url":"#b-06ae2b95bccfea8a2eeea2d4","role":"section-title"},{"id":"b-9a306ce13fe5423c3fcd3b6d","url":"#b-9a306ce13fe5423c3fcd3b6d","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-ea6996e655977b7b3e7aae39","type":"p","scope":"retrieval-approximation-and-interaction","url":"https://codingai-lec03.pages.dev/course/blocks/retrieval-approximation-and-interaction.html#b-ea6996e655977b7b3e7aae39","content":"这部分服务于第八讲的检索预备阅读，不是 P59 必须讲完的技术章。LAB12 的 IVF-Flat 先把图库分成 8 个簇，查询只打开 `nprobe` 个最近簇，再对桶内原向量精确打分。近似出现在没有打开的桶；把所有 8 个桶打开应恢复 exact 排名。","sha256":"1766e359ec4d31dd1478ef54c245a79bf9e3cd2bcb748e45f6e24d363cc4b064","chapter":{"id":"III","title":"视觉与语言"},"unit":{"id":"retrieval-cache-and-evaluation","title":"检索到底复用了什么，又省去了什么？"},"page":{"id":"reading-opening","title":"选读：省搜索和增加图文交互，是不同方向","number":1},"sectionTitle":"选读：省搜索和增加图文交互，是不同方向","currentQuestion":"选读：省搜索和增加图文交互，是不同方向","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-9cd8158ff12624e30dd37e7c","nextBlock":"b-29527c419f693494a5470d00","prerequisites":[{"id":"b-06ae2b95bccfea8a2eeea2d4","url":"#b-06ae2b95bccfea8a2eeea2d4","role":"section-title"},{"id":"b-9a306ce13fe5423c3fcd3b6d","url":"#b-9a306ce13fe5423c3fcd3b6d","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-29527c419f693494a5470d00","type":"p","scope":"retrieval-approximation-and-interaction","url":"https://codingai-lec03.pages.dev/course/blocks/retrieval-approximation-and-interaction.html#b-29527c419f693494a5470d00","content":"2026-09-22 保存的本地结果中，nprobe=1 平均扫描 4.17 个图像向量，exact 扫描 24 个；但整批 12 查询中位耗时为约 0.1016 ms 与 0.0456 ms。小库矩阵乘法已经很便宜，Python 路由和排序成本反而更大。扫描更少不等于实测更快。计时排除模型编码、磁盘读取和建索引。","sha256":"dfb749be9ea401298591c13840469fdf0010697e73c244978111ebf419a560a3","chapter":{"id":"III","title":"视觉与语言"},"unit":{"id":"retrieval-cache-and-evaluation","title":"检索到底复用了什么，又省去了什么？"},"page":{"id":"reading-opening","title":"选读：省搜索和增加图文交互，是不同方向","number":1},"sectionTitle":"选读：省搜索和增加图文交互，是不同方向","currentQuestion":"选读：省搜索和增加图文交互，是不同方向","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-ea6996e655977b7b3e7aae39","nextBlock":"b-7177cf80bc859c52ca80b6b2","prerequisites":[{"id":"b-06ae2b95bccfea8a2eeea2d4","url":"#b-06ae2b95bccfea8a2eeea2d4","role":"section-title"},{"id":"b-9a306ce13fe5423c3fcd3b6d","url":"#b-9a306ce13fe5423c3fcd3b6d","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-7177cf80bc859c52ca80b6b2","type":"p","scope":"retrieval-approximation-and-interaction","url":"https://codingai-lec03.pages.dev/course/blocks/retrieval-approximation-and-interaction.html#b-7177cf80bc859c52ca80b6b2","content":"另一个方向是让查询参与局部交互。缓存左右区域向量 $r_j$，对查询 $q$ 计算 $u_j=q^Tr_j$、$w_j=\\mathrm{softmax}(u_j/.05)$，再汇总 $\\sum_j w_ju_j$。猫查询可提高猫区域权重，宇航员查询可提高宇航员区域权重；但这个聚合对区域顺序不敏感，交换左右后总分不变。多向量并不自动保存关系。","sha256":"6424f8550f5b5c6628f1965c5c1b4aafa939764425130e58f1a53d555fcca12a","chapter":{"id":"III","title":"视觉与语言"},"unit":{"id":"retrieval-cache-and-evaluation","title":"检索到底复用了什么，又省去了什么？"},"page":{"id":"reading-opening","title":"选读：省搜索和增加图文交互，是不同方向","number":1},"sectionTitle":"选读：省搜索和增加图文交互，是不同方向","currentQuestion":"选读：省搜索和增加图文交互，是不同方向","sourceSlides":[],"knowledgeBlock":null,"formulas":[{"latex":"r_j","definitionContext":"选读：省搜索和增加图文交互，是不同方向"},{"latex":"q","definitionContext":"选读：省搜索和增加图文交互，是不同方向"},{"latex":"u_j=q^Tr_j","definitionContext":"选读：省搜索和增加图文交互，是不同方向"},{"latex":"w_j=\\mathrm{softmax}(u_j/.05)","definitionContext":"选读：省搜索和增加图文交互，是不同方向"},{"latex":"\\sum_j w_ju_j","definitionContext":"选读：省搜索和增加图文交互，是不同方向"}],"previousBlock":"b-29527c419f693494a5470d00","nextBlock":"b-686882a361bdee07a6ced45b","prerequisites":[{"id":"b-06ae2b95bccfea8a2eeea2d4","url":"#b-06ae2b95bccfea8a2eeea2d4","role":"section-title"},{"id":"b-9a306ce13fe5423c3fcd3b6d","url":"#b-9a306ce13fe5423c3fcd3b6d","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-686882a361bdee07a6ced45b","type":"p","scope":"retrieval-approximation-and-interaction","url":"https://codingai-lec03.pages.dev/course/blocks/retrieval-approximation-and-interaction.html#b-686882a361bdee07a6ced45b","content":"逐对重复调用独立 CLIP 编码器仍是 $f(I)^Tg(T)$，不因此变成交叉编码。联合模型把图文在内部相互条件化，通常需要更多查询相关计算。可用双编码先召回，再少量重排；要判断收益，需要固定检索库、查询和评价协议。这些机制在现有 LAB12 保留为选做，不在主讲另开一章。","sha256":"4317d93511e6997230f0f90631867ff02fa98c7c130893b0bfe4a4e45f7238d1","chapter":{"id":"III","title":"视觉与语言"},"unit":{"id":"retrieval-cache-and-evaluation","title":"检索到底复用了什么，又省去了什么？"},"page":{"id":"reading-opening","title":"选读：省搜索和增加图文交互，是不同方向","number":1},"sectionTitle":"选读：省搜索和增加图文交互，是不同方向","currentQuestion":"选读：省搜索和增加图文交互，是不同方向","sourceSlides":[],"knowledgeBlock":null,"formulas":[{"latex":"f(I)^Tg(T)","definitionContext":"选读：省搜索和增加图文交互，是不同方向"}],"previousBlock":"b-7177cf80bc859c52ca80b6b2","nextBlock":"b-76b28b9ffc1b109f32ac949e","prerequisites":[{"id":"b-06ae2b95bccfea8a2eeea2d4","url":"#b-06ae2b95bccfea8a2eeea2d4","role":"section-title"},{"id":"b-9a306ce13fe5423c3fcd3b6d","url":"#b-9a306ce13fe5423c3fcd3b6d","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-76b28b9ffc1b109f32ac949e","type":"h2","scope":"retrieval-approximation-and-interaction","url":"https://codingai-lec03.pages.dev/course/blocks/retrieval-approximation-and-interaction.html#b-76b28b9ffc1b109f32ac949e","content":"## 继续阅读与实验","sha256":"2b9ccef7e1220e2250cceff1abd8bdd736b42f8da5a223f26b029f44b962e807","chapter":{"id":"III","title":"视觉与语言"},"unit":{"id":"retrieval-cache-and-evaluation","title":"检索到底复用了什么，又省去了什么？"},"page":{"id":"reading-opening","title":"选读：省搜索和增加图文交互，是不同方向","number":1},"sectionTitle":"选读：省搜索和增加图文交互，是不同方向","currentQuestion":"选读：省搜索和增加图文交互，是不同方向","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-686882a361bdee07a6ced45b","nextBlock":null,"prerequisites":[{"id":"b-06ae2b95bccfea8a2eeea2d4","url":"#b-06ae2b95bccfea8a2eeea2d4","role":"section-title"},{"id":"b-9a306ce13fe5423c3fcd3b6d","url":"#b-9a306ce13fe5423c3fcd3b6d","role":"problem-context"}],"relatedBlocks":[]}]}
