{"version":2,"relationshipURLBase":"https://codingai-lec03.pages.dev/course/blocks/clip-positive-transfer-evidence.html","relationshipResolution":"Resolve relative relationship URLs against relationshipURLBase; complete records and absolute URLs are in blocks.","title":"正面证据：自然语言监督确实拓宽了下游接口","url":"https://codingai-lec03.pages.dev/course/blocks/clip-positive-transfer-evidence.html","markdown":"https://codingai-lec03.pages.dev/course/blocks/clip-positive-transfer-evidence.md","pages":[{"id":"reading-opening","title":"正面证据：自然语言监督确实拓宽了下游接口","unit":"规模化图文监督带来了哪些可验证的进展？","unitId":"clip-data-and-transfer","chapter":"III","number":1}],"aliases":[{"oldId":"b-c2d6252008e512f1a1a77c7e","targetId":"b-c2d6252008e512f1a1a77c7e","page":"reading-opening","operation":"move","beforeSHA256":"8f82ef2096e1f210f4d2eebbbe62004ab461108bff26cad87bacc6a3e4bb693f","afterSHA256":"8f82ef2096e1f210f4d2eebbbe62004ab461108bff26cad87bacc6a3e4bb693f","reason":"Same source body in authored semantic page"},{"oldId":"b-138ecaba891a99fc9f1eb723","targetId":"b-138ecaba891a99fc9f1eb723","page":"reading-opening","operation":"move","beforeSHA256":"0d653233ba094cef5ff2f1272d447699b1f92f65dbfb71291aa002592f5e7e70","afterSHA256":"0d653233ba094cef5ff2f1272d447699b1f92f65dbfb71291aa002592f5e7e70","reason":"Same source body in authored semantic page"},{"oldId":"b-708e91aee937cdf3cc5f863f","targetId":"b-708e91aee937cdf3cc5f863f","page":"reading-opening","operation":"move","beforeSHA256":"6d7f0dd74b4667c6a08b0646dce855f483d825be4b813261f23dc52e0afb3b1f","afterSHA256":"6d7f0dd74b4667c6a08b0646dce855f483d825be4b813261f23dc52e0afb3b1f","reason":"Same source body in authored semantic page"},{"oldId":"b-b24db3edcd819d291623aafc","targetId":"b-b24db3edcd819d291623aafc","page":"reading-opening","operation":"move","beforeSHA256":"1212618a46874991135f8b45069f106d46e0cb3f6ac860805804ed731db72c91","afterSHA256":"1212618a46874991135f8b45069f106d46e0cb3f6ac860805804ed731db72c91","reason":"Same source body in authored semantic page"},{"oldId":"b-4ac41e6417729065e5d19385","targetId":"b-4ac41e6417729065e5d19385","page":"reading-opening","operation":"move","beforeSHA256":"1ac2ccaaed2600cd171432e84791b757c5c816d12897751f508cab63ff1af1dc","afterSHA256":"1ac2ccaaed2600cd171432e84791b757c5c816d12897751f508cab63ff1af1dc","reason":"Same source body in authored semantic page"},{"oldId":"b-809ddd977cca53bd439dde65","targetId":"b-809ddd977cca53bd439dde65","page":"reading-opening","operation":"move","beforeSHA256":"2b9ccef7e1220e2250cceff1abd8bdd736b42f8da5a223f26b029f44b962e807","afterSHA256":"2b9ccef7e1220e2250cceff1abd8bdd736b42f8da5a223f26b029f44b962e807","reason":"Same source body in authored semantic page"}],"blocks":[{"id":"b-c2d6252008e512f1a1a77c7e","type":"h1","scope":"clip-positive-transfer-evidence","url":"https://codingai-lec03.pages.dev/course/blocks/clip-positive-transfer-evidence.html#b-c2d6252008e512f1a1a77c7e","content":"# 正面证据：自然语言监督确实拓宽了下游接口","sha256":"8f82ef2096e1f210f4d2eebbbe62004ab461108bff26cad87bacc6a3e4bb693f","chapter":{"id":"III","title":"视觉与语言"},"unit":{"id":"clip-data-and-transfer","title":"规模化图文监督带来了哪些可验证的进展？"},"page":{"id":"reading-opening","title":"正面证据：自然语言监督确实拓宽了下游接口","number":1},"sectionTitle":"正面证据：自然语言监督确实拓宽了下游接口","currentQuestion":"正面证据：自然语言监督确实拓宽了下游接口","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":null,"nextBlock":"b-138ecaba891a99fc9f1eb723","prerequisites":[{"id":"b-138ecaba891a99fc9f1eb723","url":"#b-138ecaba891a99fc9f1eb723","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-138ecaba891a99fc9f1eb723","type":"p","scope":"clip-positive-transfer-evidence","url":"https://codingai-lec03.pages.dev/course/blocks/clip-positive-transfer-evidence.html#b-138ecaba891a99fc9f1eb723","content":"CLIP 选择配对识别作为学习任务。两套编码器各自处理图片和文字，配对资料指出哪些对应应更容易被选出。与图像自监督对比的联系是训练关系的来源不同；文字监督拓展语义供给，下一页才用矩阵落实损失。","sha256":"0d653233ba094cef5ff2f1272d447699b1f92f65dbfb71291aa002592f5e7e70","chapter":{"id":"III","title":"视觉与语言"},"unit":{"id":"clip-data-and-transfer","title":"规模化图文监督带来了哪些可验证的进展？"},"page":{"id":"reading-opening","title":"正面证据：自然语言监督确实拓宽了下游接口","number":1},"sectionTitle":"正面证据：自然语言监督确实拓宽了下游接口","currentQuestion":"正面证据：自然语言监督确实拓宽了下游接口","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-c2d6252008e512f1a1a77c7e","nextBlock":"b-708e91aee937cdf3cc5f863f","prerequisites":[{"id":"b-c2d6252008e512f1a1a77c7e","url":"#b-c2d6252008e512f1a1a77c7e","role":"section-title"}],"relatedBlocks":[]},{"id":"b-708e91aee937cdf3cc5f863f","type":"p","scope":"clip-positive-transfer-evidence","url":"https://codingai-lec03.pages.dev/course/blocks/clip-positive-transfer-evidence.html#b-708e91aee937cdf3cc5f863f","content":"CLIP 的论文贡献不只是一种损失公式。原工作研究了 30 多个视觉数据集上的迁移，且报告其 zero-shot 模型在 ImageNet 上达到原始 ResNet-50 的分类准确率水平，而不使用该任务的 128 万张训练样本训练专用分类器。数字与比较属于原论文协议，不能从本课程猫图实验重新推出。","sha256":"6d7f0dd74b4667c6a08b0646dce855f483d825be4b813261f23dc52e0afb3b1f","chapter":{"id":"III","title":"视觉与语言"},"unit":{"id":"clip-data-and-transfer","title":"规模化图文监督带来了哪些可验证的进展？"},"page":{"id":"reading-opening","title":"正面证据：自然语言监督确实拓宽了下游接口","number":1},"sectionTitle":"正面证据：自然语言监督确实拓宽了下游接口","currentQuestion":"正面证据：自然语言监督确实拓宽了下游接口","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-138ecaba891a99fc9f1eb723","nextBlock":"b-b24db3edcd819d291623aafc","prerequisites":[{"id":"b-c2d6252008e512f1a1a77c7e","url":"#b-c2d6252008e512f1a1a77c7e","role":"section-title"},{"id":"b-138ecaba891a99fc9f1eb723","url":"#b-138ecaba891a99fc9f1eb723","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-b24db3edcd819d291623aafc","type":"p","scope":"clip-positive-transfer-evidence","url":"https://codingai-lec03.pages.dev/course/blocks/clip-positive-transfer-evidence.html#b-b24db3edcd819d291623aafc","content":"这项结果支持一个明确判断：大规模图文匹配学到的表示，可以通过文字描述定义一批下游分类任务，减少逐任务重新标注和训练固定分类头的需求。语言成为任务指定接口，是对“分类表只能在训练时固定”的实质拓展。","sha256":"1212618a46874991135f8b45069f106d46e0cb3f6ac860805804ed731db72c91","chapter":{"id":"III","title":"视觉与语言"},"unit":{"id":"clip-data-and-transfer","title":"规模化图文监督带来了哪些可验证的进展？"},"page":{"id":"reading-opening","title":"正面证据：自然语言监督确实拓宽了下游接口","number":1},"sectionTitle":"正面证据：自然语言监督确实拓宽了下游接口","currentQuestion":"正面证据：自然语言监督确实拓宽了下游接口","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-708e91aee937cdf3cc5f863f","nextBlock":"b-4ac41e6417729065e5d19385","prerequisites":[{"id":"b-c2d6252008e512f1a1a77c7e","url":"#b-c2d6252008e512f1a1a77c7e","role":"section-title"},{"id":"b-138ecaba891a99fc9f1eb723","url":"#b-138ecaba891a99fc9f1eb723","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-4ac41e6417729065e5d19385","type":"p","scope":"clip-positive-transfer-evidence","url":"https://codingai-lec03.pages.dev/course/blocks/clip-positive-transfer-evidence.html#b-4ac41e6417729065e5d19385","content":"然后再问边界才有意义：候选描述写得好不好、预训练是否覆盖概念、目标数据分布是否变化，都会影响迁移。关系定位和自由生成还要求别的输出。不是因为 CLIP 什么也没有证明而要继续研究，而是它已经打开了可用接口，研究者因此可以追问更细的空间、关系和交互能力。","sha256":"1ac2ccaaed2600cd171432e84791b757c5c816d12897751f508cab63ff1af1dc","chapter":{"id":"III","title":"视觉与语言"},"unit":{"id":"clip-data-and-transfer","title":"规模化图文监督带来了哪些可验证的进展？"},"page":{"id":"reading-opening","title":"正面证据：自然语言监督确实拓宽了下游接口","number":1},"sectionTitle":"正面证据：自然语言监督确实拓宽了下游接口","currentQuestion":"正面证据：自然语言监督确实拓宽了下游接口","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-b24db3edcd819d291623aafc","nextBlock":"b-809ddd977cca53bd439dde65","prerequisites":[{"id":"b-c2d6252008e512f1a1a77c7e","url":"#b-c2d6252008e512f1a1a77c7e","role":"section-title"},{"id":"b-138ecaba891a99fc9f1eb723","url":"#b-138ecaba891a99fc9f1eb723","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-809ddd977cca53bd439dde65","type":"h2","scope":"clip-positive-transfer-evidence","url":"https://codingai-lec03.pages.dev/course/blocks/clip-positive-transfer-evidence.html#b-809ddd977cca53bd439dde65","content":"## 继续阅读与实验","sha256":"2b9ccef7e1220e2250cceff1abd8bdd736b42f8da5a223f26b029f44b962e807","chapter":{"id":"III","title":"视觉与语言"},"unit":{"id":"clip-data-and-transfer","title":"规模化图文监督带来了哪些可验证的进展？"},"page":{"id":"reading-opening","title":"正面证据：自然语言监督确实拓宽了下游接口","number":1},"sectionTitle":"正面证据：自然语言监督确实拓宽了下游接口","currentQuestion":"正面证据：自然语言监督确实拓宽了下游接口","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-4ac41e6417729065e5d19385","nextBlock":null,"prerequisites":[{"id":"b-c2d6252008e512f1a1a77c7e","url":"#b-c2d6252008e512f1a1a77c7e","role":"section-title"},{"id":"b-138ecaba891a99fc9f1eb723","url":"#b-138ecaba891a99fc9f1eb723","role":"problem-context"}],"relatedBlocks":[]}]}