{"version":2,"relationshipURLBase":"https://codingai-lec03.pages.dev/course/notebook-blocks/a7a8f9c43bc64c6e8609e87186f7cb77.html","relationshipResolution":"Resolve relative relationship URLs against relationshipURLBase; complete records and absolute URLs are in blocks.","title":"从两张视图到一个可微分的选择题","url":"https://codingai-lec03.pages.dev/course/notebook-blocks/a7a8f9c43bc64c6e8609e87186f7cb77.html","markdown":"https://codingai-lec03.pages.dev/course/notebook-blocks/a7a8f9c43bc64c6e8609e87186f7cb77.md","pages":[{"id":"reading-opening","title":"从两张视图到一个可微分的选择题","unit":"从两张视图到一个可微分的选择题","unitId":"notebook-blocks/a7a8f9c43bc64c6e8609e87186f7cb77","chapter":"course","number":1}],"aliases":[{"oldId":"b-36a4e12bf1c1cf36f1a16531","targetId":"b-36a4e12bf1c1cf36f1a16531","page":"reading-opening","operation":"move","beforeSHA256":"406a009759933f5c21b004890953083cbb3ecda9532cde4f1ef9551b2f5d2da5","afterSHA256":"406a009759933f5c21b004890953083cbb3ecda9532cde4f1ef9551b2f5d2da5","reason":"Same source body in authored semantic page"},{"oldId":"b-f530efb64024818fbb02c809","targetId":"b-f530efb64024818fbb02c809","page":"reading-opening","operation":"move","beforeSHA256":"af29143428aeacff20eaa524d22ce7bb550c7fdcdd50360a9a95742c61c5e634","afterSHA256":"af29143428aeacff20eaa524d22ce7bb550c7fdcdd50360a9a95742c61c5e634","reason":"Same source body in authored semantic page"},{"oldId":"b-d350a83ea72ec4f6af15a8da","targetId":"b-d350a83ea72ec4f6af15a8da","page":"reading-opening","operation":"move","beforeSHA256":"e91f24e4e4ca40e2099a0d81d30db6568f4bf2074b499eeee6e3959617bd97ac","afterSHA256":"e91f24e4e4ca40e2099a0d81d30db6568f4bf2074b499eeee6e3959617bd97ac","reason":"Same source body in authored semantic page"},{"oldId":"b-936e36ca4fb1011f0d67272a","targetId":"b-936e36ca4fb1011f0d67272a","page":"reading-opening","operation":"move","beforeSHA256":"c023933b2170d7e6bbe13f73a3029a2b1c1858f66f78f1255a4634227653e856","afterSHA256":"c023933b2170d7e6bbe13f73a3029a2b1c1858f66f78f1255a4634227653e856","reason":"Same source body in authored semantic page"},{"oldId":"b-6c70ea8e44cdc91fb7b85430","targetId":"b-6c70ea8e44cdc91fb7b85430","page":"reading-opening","operation":"move","beforeSHA256":"9412c334ba69cecafc029ab0c22814d9df928a6c4d41cd69390e96b2b55ccfe2","afterSHA256":"9412c334ba69cecafc029ab0c22814d9df928a6c4d41cd69390e96b2b55ccfe2","reason":"Same source body in authored semantic page"},{"oldId":"b-7816726380775c09835696a0","targetId":"b-7816726380775c09835696a0","page":"reading-opening","operation":"move","beforeSHA256":"9cb323f152c42ea8e4399cbe3e89d1da1eb963999f394dfd495a7f81ec530058","afterSHA256":"9cb323f152c42ea8e4399cbe3e89d1da1eb963999f394dfd495a7f81ec530058","reason":"Same source body in authored semantic page"},{"oldId":"b-9fa5de9d24bb09e5ec4af761","targetId":"b-9fa5de9d24bb09e5ec4af761","page":"reading-opening","operation":"move","beforeSHA256":"f935511521291987e8d32a9d2f118aead493bf7ee5d87ccef6005b301011abf6","afterSHA256":"f935511521291987e8d32a9d2f118aead493bf7ee5d87ccef6005b301011abf6","reason":"Same source body in authored semantic page"},{"oldId":"b-a6ea5b44fce5cc70a9326476","targetId":"b-a6ea5b44fce5cc70a9326476","page":"reading-opening","operation":"move","beforeSHA256":"a14cb3bb25ac2e163235dba6292f6c6dc48dcfcf9dc59e99ae843a1fa8e2f1ed","afterSHA256":"a14cb3bb25ac2e163235dba6292f6c6dc48dcfcf9dc59e99ae843a1fa8e2f1ed","reason":"Same source body in authored semantic page"},{"oldId":"b-3fda2948a19f367a338487f8","targetId":"b-3fda2948a19f367a338487f8","page":"reading-opening","operation":"move","beforeSHA256":"9eaa2737f13a8bb403958a4b5ea1f8413c4fe21102fc1568aa8462c83c674588","afterSHA256":"9eaa2737f13a8bb403958a4b5ea1f8413c4fe21102fc1568aa8462c83c674588","reason":"Same source body in authored semantic page"},{"oldId":"b-2412670b8aefb136ddfa8aae","targetId":"b-2412670b8aefb136ddfa8aae","page":"reading-opening","operation":"move","beforeSHA256":"f28b0a52119137411dd8ab12c666dd9da5e57c4838b284b24b74ebffb6976cfe","afterSHA256":"f28b0a52119137411dd8ab12c666dd9da5e57c4838b284b24b74ebffb6976cfe","reason":"Same source body in authored semantic page"},{"oldId":"b-3989962ff2516ea101959383","targetId":"b-3989962ff2516ea101959383","page":"reading-opening","operation":"move","beforeSHA256":"e8ad5b16d365d3ed5a5c7ce3fde725c74331832f6e62b2f7bf33fd1909200cfd","afterSHA256":"e8ad5b16d365d3ed5a5c7ce3fde725c74331832f6e62b2f7bf33fd1909200cfd","reason":"Same source body in authored semantic page"},{"oldId":"b-fce14bcb5f878faa28eb17f5","targetId":"b-fce14bcb5f878faa28eb17f5","page":"reading-opening","operation":"move","beforeSHA256":"9199b465c0c665ebce4759b252fe603a1a408fbfae1bb3ab94339ee2add2c802","afterSHA256":"9199b465c0c665ebce4759b252fe603a1a408fbfae1bb3ab94339ee2add2c802","reason":"Same source body in authored semantic page"},{"oldId":"b-15e0e0eb68bd5fe4540aaf6f","targetId":"b-15e0e0eb68bd5fe4540aaf6f","page":"reading-opening","operation":"move","beforeSHA256":"adc84daeb588357c3c31e6da7d6c09ebecd6078e74e8e6636cf48f7a8e775f77","afterSHA256":"adc84daeb588357c3c31e6da7d6c09ebecd6078e74e8e6636cf48f7a8e775f77","reason":"Same source body in authored semantic page"},{"oldId":"b-253a0c41290554990010a9aa","targetId":"b-253a0c41290554990010a9aa","page":"reading-opening","operation":"move","beforeSHA256":"1571dbae1ada9dbf34cdbbd00400cd017e7a05cc5fc496dcd5ffe97ba7a8a7b7","afterSHA256":"1571dbae1ada9dbf34cdbbd00400cd017e7a05cc5fc496dcd5ffe97ba7a8a7b7","reason":"Same source body in authored semantic page"},{"oldId":"b-d97667c4ddf88e567d5029af","targetId":"b-d97667c4ddf88e567d5029af","page":"reading-opening","operation":"move","beforeSHA256":"fa7a35afd1de0c0e4d821a10fd183adf7b8e20a7e2bc1c49bda8198352aefe7b","afterSHA256":"fa7a35afd1de0c0e4d821a10fd183adf7b8e20a7e2bc1c49bda8198352aefe7b","reason":"Same source body in authored semantic page"},{"oldId":"b-cc6657d92ac8ee2789b80c65","targetId":"b-cc6657d92ac8ee2789b80c65","page":"reading-opening","operation":"move","beforeSHA256":"f56c232bd52e5020fa340c9c136070bd13e9ceee464f986929f7ac9606efd5c1","afterSHA256":"f56c232bd52e5020fa340c9c136070bd13e9ceee464f986929f7ac9606efd5c1","reason":"Same source body in authored semantic page"},{"oldId":"b-7c4ec1955712ec2e87d8e413","targetId":"b-7c4ec1955712ec2e87d8e413","page":"reading-opening","operation":"move","beforeSHA256":"4c3e6341c20928a51a160dedbf3f400e6361ad7a8d7eb431ea5a89a28ce1eaea","afterSHA256":"4c3e6341c20928a51a160dedbf3f400e6361ad7a8d7eb431ea5a89a28ce1eaea","reason":"Same source body in authored semantic page"},{"oldId":"b-5d92998b0c79fa82f48cdb23","targetId":"b-5d92998b0c79fa82f48cdb23","page":"reading-opening","operation":"rewrite","beforeSHA256":"51f3e5310da18b988e306f65684313e7e71e01860bb54a6a00494d8bffce1362","afterSHA256":"aee97eabc966fb919064f5873a200fd72b57e5a97fdd59c88abdc230c48e6942","reason":"Same source identity with figure semantics or reading context enriched"},{"oldId":"b-50e6354dcfc4f410e3e7bc59","targetId":"b-50e6354dcfc4f410e3e7bc59","page":"reading-opening","operation":"move","beforeSHA256":"7ba3018ebf92c10b1fc78f3cedab36def010e2813e96cfe56b3fff05bd46a085","afterSHA256":"7ba3018ebf92c10b1fc78f3cedab36def010e2813e96cfe56b3fff05bd46a085","reason":"Same source body in authored semantic page"},{"oldId":"b-e96c82ae74e84e521bf0e65b","targetId":"b-e96c82ae74e84e521bf0e65b","page":"reading-opening","operation":"move","beforeSHA256":"9ae01a37f9c65f1d1a18d40eca98d9792d0f52c55c9e1d68ba854b1f528c4cd6","afterSHA256":"9ae01a37f9c65f1d1a18d40eca98d9792d0f52c55c9e1d68ba854b1f528c4cd6","reason":"Same source body in authored semantic page"},{"oldId":"b-2351d28e0bac3cd3f0f3345c","targetId":"b-2351d28e0bac3cd3f0f3345c","page":"reading-opening","operation":"move","beforeSHA256":"60c6c757dc9996e3547d297bf1471b72cf1cb47de89e7adb3fdf2074e6660e62","afterSHA256":"60c6c757dc9996e3547d297bf1471b72cf1cb47de89e7adb3fdf2074e6660e62","reason":"Same source body in authored semantic page"},{"oldId":"b-3e9a88c5c0afec4e1aa651d8","targetId":"b-3e9a88c5c0afec4e1aa651d8","page":"reading-opening","operation":"move","beforeSHA256":"162818d06b0c1920ea801f7a914ddee919bcf66303205da1d7ecfa2c002f0e0d","afterSHA256":"162818d06b0c1920ea801f7a914ddee919bcf66303205da1d7ecfa2c002f0e0d","reason":"Same source body in authored semantic page"},{"oldId":"b-068167c9c0ee095f5734e8b1","targetId":"b-068167c9c0ee095f5734e8b1","page":"reading-opening","operation":"move","beforeSHA256":"1e9d049048183aa6e6e76b4872e6d46739728419873e176868c84cb1e0419b25","afterSHA256":"1e9d049048183aa6e6e76b4872e6d46739728419873e176868c84cb1e0419b25","reason":"Same source body in authored semantic page"},{"oldId":"b-9abe50ec4c6816bee42ebb71","targetId":"b-9abe50ec4c6816bee42ebb71","page":"reading-opening","operation":"move","beforeSHA256":"5d23cd73ca3e47dcbef8d61f23e2fa0ab3f423d612b6dd4408773577fc82375f","afterSHA256":"5d23cd73ca3e47dcbef8d61f23e2fa0ab3f423d612b6dd4408773577fc82375f","reason":"Same source body in authored semantic page"},{"oldId":"b-24730bf32aae8b178c77fcb3","targetId":"b-24730bf32aae8b178c77fcb3","page":"reading-opening","operation":"move","beforeSHA256":"0c4c6ea20d2153b2e1535a2be1939033ad5c45c42167fba81ef0c782b77d60f1","afterSHA256":"0c4c6ea20d2153b2e1535a2be1939033ad5c45c42167fba81ef0c782b77d60f1","reason":"Same source body in authored semantic page"},{"oldId":"b-be35fb878fa407db71a87d29","targetId":"b-be35fb878fa407db71a87d29","page":"reading-opening","operation":"move","beforeSHA256":"a659d64a58c06d563b62e4fe28734e91d077998a3700743c08340c069019df20","afterSHA256":"a659d64a58c06d563b62e4fe28734e91d077998a3700743c08340c069019df20","reason":"Same source body in authored semantic page"},{"oldId":"b-91682d0d3d85a1c82e5ecc04","targetId":"b-91682d0d3d85a1c82e5ecc04","page":"reading-opening","operation":"move","beforeSHA256":"542bbbbbf07478a29b46b11c0e054862781ed72fdbffa19e42b223caf08168a2","afterSHA256":"542bbbbbf07478a29b46b11c0e054862781ed72fdbffa19e42b223caf08168a2","reason":"Same source body in authored semantic page"},{"oldId":"b-6936c0a1ebe3e8f3e96aee3e","targetId":"b-6936c0a1ebe3e8f3e96aee3e","page":"reading-opening","operation":"move","beforeSHA256":"cdfb9cdcad42224287fc11f2cf2313e29b96100a33841b6a10511d0ba3b8adea","afterSHA256":"cdfb9cdcad42224287fc11f2cf2313e29b96100a33841b6a10511d0ba3b8adea","reason":"Same source body in authored semantic page"},{"oldId":"block-a7a8f9c43bc64c6e8609e87186f7cb77--从两张视图到一个可微分的选择题","targetId":"b-2412670b8aefb136ddfa8aae","operation":"retain-source-identity"},{"oldId":"block-a7a8f9c43bc64c6e8609e87186f7cb77--温度改变竞争强度不改变这组相似度的排序","targetId":"b-cc6657d92ac8ee2789b80c65","operation":"retain-source-identity"},{"oldId":"block-a7a8f9c43bc64c6e8609e87186f7cb77--为什么损失作用在-z迁移却常读-h","targetId":"b-24730bf32aae8b178c77fcb3","operation":"retain-source-identity"}],"blocks":[{"id":"b-36a4e12bf1c1cf36f1a16531","type":"p","scope":"document","url":"https://codingai-lec03.pages.dev/course/notebook-blocks/a7a8f9c43bc64c6e8609e87186f7cb77.html#b-36a4e12bf1c1cf36f1a16531","content":"[CLS 05 · 对比学习：视图、投影与历史字典](/course/notebooks/b8320404134f4a2e858ccfe7d50a71d3.html)","sha256":"406a009759933f5c21b004890953083cbb3ecda9532cde4f1ef9551b2f5d2da5","chapter":{"id":"course","title":"从两张视图到一个可微分的选择题"},"unit":{"id":"notebook-blocks/a7a8f9c43bc64c6e8609e87186f7cb77","title":"从两张视图到一个可微分的选择题"},"page":{"id":"reading-opening","title":"从两张视图到一个可微分的选择题","number":1},"sectionTitle":"从两张视图到一个可微分的选择题","currentQuestion":"从两张视图到一个可微分的选择题","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":null,"nextBlock":"b-f530efb64024818fbb02c809","prerequisites":[],"relatedBlocks":[]},{"id":"b-f530efb64024818fbb02c809","type":"h1","scope":"document","url":"https://codingai-lec03.pages.dev/course/notebook-blocks/a7a8f9c43bc64c6e8609e87186f7cb77.html#b-f530efb64024818fbb02c809","content":"# 从两张视图到一个可微分的选择题","sha256":"af29143428aeacff20eaa524d22ce7bb550c7fdcdd50360a9a95742c61c5e634","chapter":{"id":"course","title":"从两张视图到一个可微分的选择题"},"unit":{"id":"notebook-blocks/a7a8f9c43bc64c6e8609e87186f7cb77","title":"从两张视图到一个可微分的选择题"},"page":{"id":"reading-opening","title":"从两张视图到一个可微分的选择题","number":1},"sectionTitle":"从两张视图到一个可微分的选择题","currentQuestion":"从两张视图到一个可微分的选择题","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-36a4e12bf1c1cf36f1a16531","nextBlock":"b-d350a83ea72ec4f6af15a8da","prerequisites":[{"id":"b-36a4e12bf1c1cf36f1a16531","url":"#b-36a4e12bf1c1cf36f1a16531","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-d350a83ea72ec4f6af15a8da","type":"p","scope":"document","url":"https://codingai-lec03.pages.dev/course/notebook-blocks/a7a8f9c43bc64c6e8609e87186f7cb77.html#b-d350a83ea72ec4f6af15a8da","content":"按问题继续","sha256":"e91f24e4e4ca40e2099a0d81d30db6568f4bf2074b499eeee6e3959617bd97ac","chapter":{"id":"course","title":"从两张视图到一个可微分的选择题"},"unit":{"id":"notebook-blocks/a7a8f9c43bc64c6e8609e87186f7cb77","title":"从两张视图到一个可微分的选择题"},"page":{"id":"reading-opening","title":"从两张视图到一个可微分的选择题","number":1},"sectionTitle":"从两张视图到一个可微分的选择题","currentQuestion":"从两张视图到一个可微分的选择题","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-f530efb64024818fbb02c809","nextBlock":"b-936e36ca4fb1011f0d67272a","prerequisites":[{"id":"b-f530efb64024818fbb02c809","url":"#b-f530efb64024818fbb02c809","role":"section-title"},{"id":"b-36a4e12bf1c1cf36f1a16531","url":"#b-36a4e12bf1c1cf36f1a16531","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-936e36ca4fb1011f0d67272a","type":"p","scope":"document","url":"https://codingai-lec03.pages.dev/course/notebook-blocks/a7a8f9c43bc64c6e8609e87186f7cb77.html#b-936e36ca4fb1011f0d67272a","content":"InfoNCE概率与梯度、h/z、温度、队列陈旧性、实测与评价协议","sha256":"c023933b2170d7e6bbe13f73a3029a2b1c1858f66f78f1255a4634227653e856","chapter":{"id":"course","title":"从两张视图到一个可微分的选择题"},"unit":{"id":"notebook-blocks/a7a8f9c43bc64c6e8609e87186f7cb77","title":"从两张视图到一个可微分的选择题"},"page":{"id":"reading-opening","title":"从两张视图到一个可微分的选择题","number":1},"sectionTitle":"从两张视图到一个可微分的选择题","currentQuestion":"从两张视图到一个可微分的选择题","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-d350a83ea72ec4f6af15a8da","nextBlock":"b-6c70ea8e44cdc91fb7b85430","prerequisites":[{"id":"b-f530efb64024818fbb02c809","url":"#b-f530efb64024818fbb02c809","role":"section-title"},{"id":"b-36a4e12bf1c1cf36f1a16531","url":"#b-36a4e12bf1c1cf36f1a16531","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-6c70ea8e44cdc91fb7b85430","type":"p","scope":"document","url":"https://codingai-lec03.pages.dev/course/notebook-blocks/a7a8f9c43bc64c6e8609e87186f7cb77.html#b-6c70ea8e44cdc91fb7b85430","content":"**学习路径：** [视图、投影头与历史字典](/course/study/contrastive-views.html)","sha256":"9412c334ba69cecafc029ab0c22814d9df928a6c4d41cd69390e96b2b55ccfe2","chapter":{"id":"course","title":"从两张视图到一个可微分的选择题"},"unit":{"id":"notebook-blocks/a7a8f9c43bc64c6e8609e87186f7cb77","title":"从两张视图到一个可微分的选择题"},"page":{"id":"reading-opening","title":"从两张视图到一个可微分的选择题","number":1},"sectionTitle":"从两张视图到一个可微分的选择题","currentQuestion":"从两张视图到一个可微分的选择题","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-936e36ca4fb1011f0d67272a","nextBlock":"b-7816726380775c09835696a0","prerequisites":[{"id":"b-f530efb64024818fbb02c809","url":"#b-f530efb64024818fbb02c809","role":"section-title"},{"id":"b-36a4e12bf1c1cf36f1a16531","url":"#b-36a4e12bf1c1cf36f1a16531","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-7816726380775c09835696a0","type":"p","scope":"document","url":"https://codingai-lec03.pages.dev/course/notebook-blocks/a7a8f9c43bc64c6e8609e87186f7cb77.html#b-7816726380775c09835696a0","content":"[P08同源视图](/course/pages/l3-r2-p08.html) → [P12投影头](/course/pages/l3-r2-p12.html) → [P17一次队列更新](/course/pages/l3-r2-p17.html)","sha256":"9cb323f152c42ea8e4399cbe3e89d1da1eb963999f394dfd495a7f81ec530058","chapter":{"id":"course","title":"从两张视图到一个可微分的选择题"},"unit":{"id":"notebook-blocks/a7a8f9c43bc64c6e8609e87186f7cb77","title":"从两张视图到一个可微分的选择题"},"page":{"id":"reading-opening","title":"从两张视图到一个可微分的选择题","number":1},"sectionTitle":"从两张视图到一个可微分的选择题","currentQuestion":"从两张视图到一个可微分的选择题","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-6c70ea8e44cdc91fb7b85430","nextBlock":"b-9fa5de9d24bb09e5ec4af761","prerequisites":[{"id":"b-f530efb64024818fbb02c809","url":"#b-f530efb64024818fbb02c809","role":"section-title"},{"id":"b-36a4e12bf1c1cf36f1a16531","url":"#b-36a4e12bf1c1cf36f1a16531","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-9fa5de9d24bb09e5ec4af761","type":"p","scope":"document","url":"https://codingai-lec03.pages.dev/course/notebook-blocks/a7a8f9c43bc64c6e8609e87186f7cb77.html#b-9fa5de9d24bb09e5ec4af761","content":"先把一张图的两种观察放入同一候选表，算出正例概率；再追踪梯度怎样经过z回到h，以及旧key为什么需要缓慢变化的编码器。[温度的概率与梯度算例](/course/blocks/temperature-ranking-probability-gradient.html)、[队列身份](/course/blocks/moco-key-queue-identity.html)都可以独立引用。","sha256":"f935511521291987e8d32a9d2f118aead493bf7ee5d87ccef6005b301011abf6","chapter":{"id":"course","title":"从两张视图到一个可微分的选择题"},"unit":{"id":"notebook-blocks/a7a8f9c43bc64c6e8609e87186f7cb77","title":"从两张视图到一个可微分的选择题"},"page":{"id":"reading-opening","title":"从两张视图到一个可微分的选择题","number":1},"sectionTitle":"从两张视图到一个可微分的选择题","currentQuestion":"从两张视图到一个可微分的选择题","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-7816726380775c09835696a0","nextBlock":"b-a6ea5b44fce5cc70a9326476","prerequisites":[{"id":"b-f530efb64024818fbb02c809","url":"#b-f530efb64024818fbb02c809","role":"section-title"},{"id":"b-36a4e12bf1c1cf36f1a16531","url":"#b-36a4e12bf1c1cf36f1a16531","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-a6ea5b44fce5cc70a9326476","type":"p","scope":"document","url":"https://codingai-lec03.pages.dev/course/notebook-blocks/a7a8f9c43bc64c6e8609e87186f7cb77.html#b-a6ea5b44fce5cc70a9326476","content":"动手只选一个入口：想检查增强留下什么，选[LAB 06](https://deepnote.com/project/38755e4e-23f6-4f5f-8c68-8a2e4114b1c1/notebook/7c4f531113d7473abc174e2e014e1340)；想追踪真实key、FIFO和EMA，选[LAB 07](https://deepnote.com/project/38755e4e-23f6-4f5f-8c68-8a2e4114b1c1/notebook/0d1b8f6fbf6a476c93ffcc169f2ec1f4)。读完用一张图或一段trace指出输入、候选和两种参数更新，再用独立读出评价它是否对任务有用。后文同时保留论文迁移进展和小型实验的失败结果，二者的数据与协议分别写明。","sha256":"a14cb3bb25ac2e163235dba6292f6c6dc48dcfcf9dc59e99ae843a1fa8e2f1ed","chapter":{"id":"course","title":"从两张视图到一个可微分的选择题"},"unit":{"id":"notebook-blocks/a7a8f9c43bc64c6e8609e87186f7cb77","title":"从两张视图到一个可微分的选择题"},"page":{"id":"reading-opening","title":"从两张视图到一个可微分的选择题","number":1},"sectionTitle":"从两张视图到一个可微分的选择题","currentQuestion":"从两张视图到一个可微分的选择题","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-9fa5de9d24bb09e5ec4af761","nextBlock":"b-3fda2948a19f367a338487f8","prerequisites":[{"id":"b-f530efb64024818fbb02c809","url":"#b-f530efb64024818fbb02c809","role":"section-title"},{"id":"b-36a4e12bf1c1cf36f1a16531","url":"#b-36a4e12bf1c1cf36f1a16531","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-3fda2948a19f367a338487f8","type":"p","scope":"document","url":"https://codingai-lec03.pages.dev/course/notebook-blocks/a7a8f9c43bc64c6e8609e87186f7cb77.html#b-3fda2948a19f367a338487f8","content":"拍同一只杯子，改变光照、裁剪和背景，类别仍然是杯子。若能把这些变化视为同一对象的不同观察，训练目标就可以来自图像自身：两次观察的表示应该接近，其他图像的表示提供区分参照。但如果下一项任务是分辨红杯和蓝杯，“改变颜色仍然算同一个”就不再无害。**对比学习的正例关系，是设计者通过数据和增强作出的任务假设。** 它不是无标签数据自动告诉我们的普遍相同性。","sha256":"9eaa2737f13a8bb403958a4b5ea1f8413c4fe21102fc1568aa8462c83c674588","chapter":{"id":"course","title":"从两张视图到一个可微分的选择题"},"unit":{"id":"notebook-blocks/a7a8f9c43bc64c6e8609e87186f7cb77","title":"从两张视图到一个可微分的选择题"},"page":{"id":"reading-opening","title":"从两张视图到一个可微分的选择题","number":1},"sectionTitle":"从两张视图到一个可微分的选择题","currentQuestion":"从两张视图到一个可微分的选择题","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-a6ea5b44fce5cc70a9326476","nextBlock":"b-2412670b8aefb136ddfa8aae","prerequisites":[{"id":"b-f530efb64024818fbb02c809","url":"#b-f530efb64024818fbb02c809","role":"section-title"},{"id":"b-36a4e12bf1c1cf36f1a16531","url":"#b-36a4e12bf1c1cf36f1a16531","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-2412670b8aefb136ddfa8aae","type":"h2","scope":"document","url":"https://codingai-lec03.pages.dev/course/notebook-blocks/a7a8f9c43bc64c6e8609e87186f7cb77.html#b-2412670b8aefb136ddfa8aae","content":"## 从两张视图到一个可微分的选择题","sha256":"f28b0a52119137411dd8ab12c666dd9da5e57c4838b284b24b74ebffb6976cfe","chapter":{"id":"course","title":"从两张视图到一个可微分的选择题"},"unit":{"id":"notebook-blocks/a7a8f9c43bc64c6e8609e87186f7cb77","title":"从两张视图到一个可微分的选择题"},"page":{"id":"reading-opening","title":"从两张视图到一个可微分的选择题","number":1},"sectionTitle":"从两张视图到一个可微分的选择题","currentQuestion":"从两张视图到一个可微分的选择题","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-3fda2948a19f367a338487f8","nextBlock":"b-3989962ff2516ea101959383","prerequisites":[{"id":"b-f530efb64024818fbb02c809","url":"#b-f530efb64024818fbb02c809","role":"section-title"},{"id":"b-36a4e12bf1c1cf36f1a16531","url":"#b-36a4e12bf1c1cf36f1a16531","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-3989962ff2516ea101959383","type":"p","scope":"document","url":"https://codingai-lec03.pages.dev/course/notebook-blocks/a7a8f9c43bc64c6e8609e87186f7cb77.html#b-3989962ff2516ea101959383","content":"给原图 $x$ 独立采样增强 $t,t'$，得到 $v=t(x),v^+=t'(x)$。编码器 $f_\\theta$ 产生表示 $h=f_\\theta(v)$；投影头 $g_\\phi$ 产生 $z=g_\\phi(h)$，通常再做 $\\ell_2$ 归一化。余弦相似度便是单位向量点积 $s(q,k)=q^Tk$。正例 key 来自同一原图另一视图，候选负例来自其他原图。","sha256":"e8ad5b16d365d3ed5a5c7ce3fde725c74331832f6e62b2f7bf33fd1909200cfd","chapter":{"id":"course","title":"从两张视图到一个可微分的选择题"},"unit":{"id":"notebook-blocks/a7a8f9c43bc64c6e8609e87186f7cb77","title":"从两张视图到一个可微分的选择题"},"page":{"id":"reading-opening","title":"从两张视图到一个可微分的选择题","number":1},"sectionTitle":"从两张视图到一个可微分的选择题","currentQuestion":"从两张视图到一个可微分的选择题","sourceSlides":[],"knowledgeBlock":null,"formulas":[{"latex":"x","definitionContext":"从两张视图到一个可微分的选择题"},{"latex":"t,t'","definitionContext":"从两张视图到一个可微分的选择题"},{"latex":"v=t(x),v^+=t'(x)","definitionContext":"从两张视图到一个可微分的选择题"},{"latex":"f_\\theta","definitionContext":"从两张视图到一个可微分的选择题"},{"latex":"h=f_\\theta(v)","definitionContext":"从两张视图到一个可微分的选择题"},{"latex":"g_\\phi","definitionContext":"从两张视图到一个可微分的选择题"},{"latex":"z=g_\\phi(h)","definitionContext":"从两张视图到一个可微分的选择题"},{"latex":"\\ell_2","definitionContext":"从两张视图到一个可微分的选择题"},{"latex":"s(q,k)=q^Tk","definitionContext":"从两张视图到一个可微分的选择题"}],"previousBlock":"b-2412670b8aefb136ddfa8aae","nextBlock":"b-fce14bcb5f878faa28eb17f5","prerequisites":[{"id":"b-2412670b8aefb136ddfa8aae","url":"#b-2412670b8aefb136ddfa8aae","role":"section-title"},{"id":"b-36a4e12bf1c1cf36f1a16531","url":"#b-36a4e12bf1c1cf36f1a16531","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-fce14bcb5f878faa28eb17f5","type":"p","scope":"document","url":"https://codingai-lec03.pages.dev/course/notebook-blocks/a7a8f9c43bc64c6e8609e87186f7cb77.html#b-fce14bcb5f878faa28eb17f5","content":"对一个 query $q$、一个正例 $k^+$ 和若干负例 $k_j^-$，写出 InfoNCE 形式：","sha256":"9199b465c0c665ebce4759b252fe603a1a408fbfae1bb3ab94339ee2add2c802","chapter":{"id":"course","title":"从两张视图到一个可微分的选择题"},"unit":{"id":"notebook-blocks/a7a8f9c43bc64c6e8609e87186f7cb77","title":"从两张视图到一个可微分的选择题"},"page":{"id":"reading-opening","title":"从两张视图到一个可微分的选择题","number":1},"sectionTitle":"从两张视图到一个可微分的选择题","currentQuestion":"从两张视图到一个可微分的选择题","sourceSlides":[],"knowledgeBlock":null,"formulas":[{"latex":"q","definitionContext":"从两张视图到一个可微分的选择题"},{"latex":"k^+","definitionContext":"从两张视图到一个可微分的选择题"},{"latex":"k_j^-","definitionContext":"从两张视图到一个可微分的选择题"}],"previousBlock":"b-3989962ff2516ea101959383","nextBlock":"b-15e0e0eb68bd5fe4540aaf6f","prerequisites":[{"id":"b-2412670b8aefb136ddfa8aae","url":"#b-2412670b8aefb136ddfa8aae","role":"section-title"},{"id":"b-36a4e12bf1c1cf36f1a16531","url":"#b-36a4e12bf1c1cf36f1a16531","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-15e0e0eb68bd5fe4540aaf6f","type":"p","scope":"document","url":"https://codingai-lec03.pages.dev/course/notebook-blocks/a7a8f9c43bc64c6e8609e87186f7cb77.html#b-15e0e0eb68bd5fe4540aaf6f","content":"$L_q=-\\log\\frac{\\exp(s(q,k^+)/\\tau)}{\\exp(s(q,k^+)/\\tau)+\\sum_j\\exp(s(q,k_j^-)/\\tau)}.$","sha256":"adc84daeb588357c3c31e6da7d6c09ebecd6078e74e8e6636cf48f7a8e775f77","chapter":{"id":"course","title":"从两张视图到一个可微分的选择题"},"unit":{"id":"notebook-blocks/a7a8f9c43bc64c6e8609e87186f7cb77","title":"从两张视图到一个可微分的选择题"},"page":{"id":"reading-opening","title":"从两张视图到一个可微分的选择题","number":1},"sectionTitle":"从两张视图到一个可微分的选择题","currentQuestion":"从两张视图到一个可微分的选择题","sourceSlides":[],"knowledgeBlock":null,"formulas":[{"latex":"L_q=-\\log\\frac{\\exp(s(q,k^+)/\\tau)}{\\exp(s(q,k^+)/\\tau)+\\sum_j\\exp(s(q,k_j^-)/\\tau)}.","definitionContext":"从两张视图到一个可微分的选择题"}],"previousBlock":"b-fce14bcb5f878faa28eb17f5","nextBlock":"b-253a0c41290554990010a9aa","prerequisites":[{"id":"b-2412670b8aefb136ddfa8aae","url":"#b-2412670b8aefb136ddfa8aae","role":"section-title"},{"id":"b-36a4e12bf1c1cf36f1a16531","url":"#b-36a4e12bf1c1cf36f1a16531","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-253a0c41290554990010a9aa","type":"p","scope":"document","url":"https://codingai-lec03.pages.dev/course/notebook-blocks/a7a8f9c43bc64c6e8609e87186f7cb77.html#b-253a0c41290554990010a9aa","content":"这里 $\\tau>0$ 是温度，分母包含正例。它把“拉近正例、推开负例”写成在当前候选集合里选中正例的交叉熵。候选身份来自采样过程，不是人工类别；两张不同原图可能都是杯子，所以负例并不保证语义类别不同。实例区分与类别聚合之间存在真实张力。","sha256":"1571dbae1ada9dbf34cdbbd00400cd017e7a05cc5fc496dcd5ffe97ba7a8a7b7","chapter":{"id":"course","title":"从两张视图到一个可微分的选择题"},"unit":{"id":"notebook-blocks/a7a8f9c43bc64c6e8609e87186f7cb77","title":"从两张视图到一个可微分的选择题"},"page":{"id":"reading-opening","title":"从两张视图到一个可微分的选择题","number":1},"sectionTitle":"从两张视图到一个可微分的选择题","currentQuestion":"从两张视图到一个可微分的选择题","sourceSlides":[],"knowledgeBlock":null,"formulas":[{"latex":"\\tau>0","definitionContext":"从两张视图到一个可微分的选择题"}],"previousBlock":"b-15e0e0eb68bd5fe4540aaf6f","nextBlock":"b-d97667c4ddf88e567d5029af","prerequisites":[{"id":"b-2412670b8aefb136ddfa8aae","url":"#b-2412670b8aefb136ddfa8aae","role":"section-title"},{"id":"b-36a4e12bf1c1cf36f1a16531","url":"#b-36a4e12bf1c1cf36f1a16531","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-d97667c4ddf88e567d5029af","type":"p","scope":"document","url":"https://codingai-lec03.pages.dev/course/notebook-blocks/a7a8f9c43bc64c6e8609e87186f7cb77.html#b-d97667c4ddf88e567d5029af","content":"SimCLR 的一个 batch 有 $N$ 张原图，生成 $2N$ 个视图。对其中任一视图，排除自身，余下 $2N-1$ 项中有一个正例和 $2N-2$ 个负例。自身相似度通常接近 1，若误把它留进分母，就额外加入一个不可避免的强竞争者。损失平均两个方向：视图 A 找 B，B 也找 A。可核对 [SimCLR §2、式 1 与 Algorithm 1](https://proceedings.mlr.press/v119/chen20j/chen20j.pdf)。","sha256":"fa7a35afd1de0c0e4d821a10fd183adf7b8e20a7e2bc1c49bda8198352aefe7b","chapter":{"id":"course","title":"从两张视图到一个可微分的选择题"},"unit":{"id":"notebook-blocks/a7a8f9c43bc64c6e8609e87186f7cb77","title":"从两张视图到一个可微分的选择题"},"page":{"id":"reading-opening","title":"从两张视图到一个可微分的选择题","number":1},"sectionTitle":"从两张视图到一个可微分的选择题","currentQuestion":"从两张视图到一个可微分的选择题","sourceSlides":[],"knowledgeBlock":null,"formulas":[{"latex":"N","definitionContext":"从两张视图到一个可微分的选择题"},{"latex":"2N","definitionContext":"从两张视图到一个可微分的选择题"},{"latex":"2N-1","definitionContext":"从两张视图到一个可微分的选择题"},{"latex":"2N-2","definitionContext":"从两张视图到一个可微分的选择题"}],"previousBlock":"b-253a0c41290554990010a9aa","nextBlock":"b-cc6657d92ac8ee2789b80c65","prerequisites":[{"id":"b-2412670b8aefb136ddfa8aae","url":"#b-2412670b8aefb136ddfa8aae","role":"section-title"},{"id":"b-36a4e12bf1c1cf36f1a16531","url":"#b-36a4e12bf1c1cf36f1a16531","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-cc6657d92ac8ee2789b80c65","type":"h2","scope":"document","url":"https://codingai-lec03.pages.dev/course/notebook-blocks/a7a8f9c43bc64c6e8609e87186f7cb77.html#b-cc6657d92ac8ee2789b80c65","content":"## 温度改变竞争强度，不改变这组相似度的排序","sha256":"f56c232bd52e5020fa340c9c136070bd13e9ceee464f986929f7ac9606efd5c1","chapter":{"id":"course","title":"从两张视图到一个可微分的选择题"},"unit":{"id":"notebook-blocks/a7a8f9c43bc64c6e8609e87186f7cb77","title":"从两张视图到一个可微分的选择题"},"page":{"id":"reading-opening","title":"从两张视图到一个可微分的选择题","number":1},"sectionTitle":"从两张视图到一个可微分的选择题","currentQuestion":"从两张视图到一个可微分的选择题","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-d97667c4ddf88e567d5029af","nextBlock":"b-7c4ec1955712ec2e87d8e413","prerequisites":[{"id":"b-2412670b8aefb136ddfa8aae","url":"#b-2412670b8aefb136ddfa8aae","role":"section-title"},{"id":"b-36a4e12bf1c1cf36f1a16531","url":"#b-36a4e12bf1c1cf36f1a16531","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-7c4ec1955712ec2e87d8e413","type":"p","scope":"document","url":"https://codingai-lec03.pages.dev/course/notebook-blocks/a7a8f9c43bc64c6e8609e87186f7cb77.html#b-7c4ec1955712ec2e87d8e413","content":"固定三项相似度：正例 0.8，两个负例 0.4、0.2。直接代入上式得到：","sha256":"4c3e6341c20928a51a160dedbf3f400e6361ad7a8d7eb431ea5a89a28ce1eaea","chapter":{"id":"course","title":"从两张视图到一个可微分的选择题"},"unit":{"id":"notebook-blocks/a7a8f9c43bc64c6e8609e87186f7cb77","title":"从两张视图到一个可微分的选择题"},"page":{"id":"reading-opening","title":"从两张视图到一个可微分的选择题","number":1},"sectionTitle":"从两张视图到一个可微分的选择题","currentQuestion":"从两张视图到一个可微分的选择题","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-cc6657d92ac8ee2789b80c65","nextBlock":"b-5d92998b0c79fa82f48cdb23","prerequisites":[{"id":"b-cc6657d92ac8ee2789b80c65","url":"#b-cc6657d92ac8ee2789b80c65","role":"section-title"},{"id":"b-36a4e12bf1c1cf36f1a16531","url":"#b-36a4e12bf1c1cf36f1a16531","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-5d92998b0c79fa82f48cdb23","type":"table","scope":"document","url":"https://codingai-lec03.pages.dev/course/notebook-blocks/a7a8f9c43bc64c6e8609e87186f7cb77.html#b-5d92998b0c79fa82f48cdb23","content":"| 温度 $\\tau$ | 正例概率 | 负例 0.4 的概率 | 负例 0.2 的概率 | 损失 $-\\log p_+$ |\n| --- | --- | --- | --- | --- |\n| 1.0 | 0.4506 | 0.3021 | 0.2473 | 0.7971 |\n| 0.5 | 0.5713 | 0.2567 | 0.1721 | 0.5599 |\n| 0.1 | 0.9796 | 0.0179 | 0.0024 | 0.0206 |","sha256":"aee97eabc966fb919064f5873a200fd72b57e5a97fdd59c88abdc230c48e6942","chapter":{"id":"course","title":"从两张视图到一个可微分的选择题"},"unit":{"id":"notebook-blocks/a7a8f9c43bc64c6e8609e87186f7cb77","title":"从两张视图到一个可微分的选择题"},"page":{"id":"reading-opening","title":"从两张视图到一个可微分的选择题","number":1},"sectionTitle":"从两张视图到一个可微分的选择题","currentQuestion":"从两张视图到一个可微分的选择题","sourceSlides":[],"knowledgeBlock":null,"formulas":[{"latex":"\\tau","definitionContext":"从两张视图到一个可微分的选择题"},{"latex":"-\\log p_+","definitionContext":"从两张视图到一个可微分的选择题"}],"table":{"headers":["温度 τ\\tauτ","正例概率","负例 0.4 的概率","负例 0.2 的概率","损失 −log⁡p+-\\log p_+−logp+​"],"rows":[["温度 τ\\tauτ","正例概率","负例 0.4 的概率","负例 0.2 的概率","损失 −log⁡p+-\\log p_+−logp+​"],["1.0","0.4506","0.3021","0.2473","0.7971"],["0.5","0.5713","0.2567","0.1721","0.5599"],["0.1","0.9796","0.0179","0.0024","0.0206"]]},"previousBlock":"b-7c4ec1955712ec2e87d8e413","nextBlock":"b-50e6354dcfc4f410e3e7bc59","prerequisites":[{"id":"b-cc6657d92ac8ee2789b80c65","url":"#b-cc6657d92ac8ee2789b80c65","role":"section-title"},{"id":"b-36a4e12bf1c1cf36f1a16531","url":"#b-36a4e12bf1c1cf36f1a16531","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-50e6354dcfc4f410e3e7bc59","type":"p","scope":"document","url":"https://codingai-lec03.pages.dev/course/notebook-blocks/a7a8f9c43bc64c6e8609e87186f7cb77.html#b-50e6354dcfc4f410e3e7bc59","content":"因为正例已经领先，低温度放大它的优势。若把正例改成 0.3、最强负例仍为 0.4，低温度则会更强地惩罚错误排序。不能仅因降低温度使这张表的损失变小，就宣称表示更好了。","sha256":"7ba3018ebf92c10b1fc78f3cedab36def010e2813e96cfe56b3fff05bd46a085","chapter":{"id":"course","title":"从两张视图到一个可微分的选择题"},"unit":{"id":"notebook-blocks/a7a8f9c43bc64c6e8609e87186f7cb77","title":"从两张视图到一个可微分的选择题"},"page":{"id":"reading-opening","title":"从两张视图到一个可微分的选择题","number":1},"sectionTitle":"从两张视图到一个可微分的选择题","currentQuestion":"从两张视图到一个可微分的选择题","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-5d92998b0c79fa82f48cdb23","nextBlock":"b-e96c82ae74e84e521bf0e65b","prerequisites":[{"id":"b-cc6657d92ac8ee2789b80c65","url":"#b-cc6657d92ac8ee2789b80c65","role":"section-title"},{"id":"b-36a4e12bf1c1cf36f1a16531","url":"#b-36a4e12bf1c1cf36f1a16531","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-e96c82ae74e84e521bf0e65b","type":"p","scope":"document","url":"https://codingai-lec03.pages.dev/course/notebook-blocks/a7a8f9c43bc64c6e8609e87186f7cb77.html#b-e96c82ae74e84e521bf0e65b","content":"梯度把这个现象写得更精确。令 $p_j$ 为候选 softmax 概率，则","sha256":"9ae01a37f9c65f1d1a18d40eca98d9792d0f52c55c9e1d68ba854b1f528c4cd6","chapter":{"id":"course","title":"从两张视图到一个可微分的选择题"},"unit":{"id":"notebook-blocks/a7a8f9c43bc64c6e8609e87186f7cb77","title":"从两张视图到一个可微分的选择题"},"page":{"id":"reading-opening","title":"从两张视图到一个可微分的选择题","number":1},"sectionTitle":"从两张视图到一个可微分的选择题","currentQuestion":"从两张视图到一个可微分的选择题","sourceSlides":[],"knowledgeBlock":null,"formulas":[{"latex":"p_j","definitionContext":"从两张视图到一个可微分的选择题"}],"previousBlock":"b-50e6354dcfc4f410e3e7bc59","nextBlock":"b-2351d28e0bac3cd3f0f3345c","prerequisites":[{"id":"b-cc6657d92ac8ee2789b80c65","url":"#b-cc6657d92ac8ee2789b80c65","role":"section-title"},{"id":"b-36a4e12bf1c1cf36f1a16531","url":"#b-36a4e12bf1c1cf36f1a16531","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-2351d28e0bac3cd3f0f3345c","type":"p","scope":"document","url":"https://codingai-lec03.pages.dev/course/notebook-blocks/a7a8f9c43bc64c6e8609e87186f7cb77.html#b-2351d28e0bac3cd3f0f3345c","content":"$\\frac{\\partial L}{\\partial s_+}=\\frac{p_+-1}{\\tau},\\qquad\n\\frac{\\partial L}{\\partial s_j^-}=\\frac{p_j}{\\tau}.$","sha256":"60c6c757dc9996e3547d297bf1471b72cf1cb47de89e7adb3fdf2074e6660e62","chapter":{"id":"course","title":"从两张视图到一个可微分的选择题"},"unit":{"id":"notebook-blocks/a7a8f9c43bc64c6e8609e87186f7cb77","title":"从两张视图到一个可微分的选择题"},"page":{"id":"reading-opening","title":"从两张视图到一个可微分的选择题","number":1},"sectionTitle":"从两张视图到一个可微分的选择题","currentQuestion":"从两张视图到一个可微分的选择题","sourceSlides":[],"knowledgeBlock":null,"formulas":[{"latex":"\\frac{\\partial L}{\\partial s_+}=\\frac{p_+-1}{\\tau},\\qquad\n\\frac{\\partial L}{\\partial s_j^-}=\\frac{p_j}{\\tau}.","definitionContext":"从两张视图到一个可微分的选择题"}],"previousBlock":"b-e96c82ae74e84e521bf0e65b","nextBlock":"b-3e9a88c5c0afec4e1aa651d8","prerequisites":[{"id":"b-cc6657d92ac8ee2789b80c65","url":"#b-cc6657d92ac8ee2789b80c65","role":"section-title"},{"id":"b-36a4e12bf1c1cf36f1a16531","url":"#b-36a4e12bf1c1cf36f1a16531","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-3e9a88c5c0afec4e1aa651d8","type":"p","scope":"document","url":"https://codingai-lec03.pages.dev/course/notebook-blocks/a7a8f9c43bc64c6e8609e87186f7cb77.html#b-3e9a88c5c0afec4e1aa651d8","content":"温度既出现在 $1/\\tau$ 中，也改变 $p$。所以“温度越低，所有梯度越大”不成立：本例正例梯度的绝对值在 $\\tau=0.5$ 时约 0.8575，在 0.1 时反而约 0.2037，因为预测已饱和。最相似负例会获得更多概率和更强排斥，但过度强调困难负例也可能放大假负例或噪声。","sha256":"162818d06b0c1920ea801f7a914ddee919bcf66303205da1d7ecfa2c002f0e0d","chapter":{"id":"course","title":"从两张视图到一个可微分的选择题"},"unit":{"id":"notebook-blocks/a7a8f9c43bc64c6e8609e87186f7cb77","title":"从两张视图到一个可微分的选择题"},"page":{"id":"reading-opening","title":"从两张视图到一个可微分的选择题","number":1},"sectionTitle":"从两张视图到一个可微分的选择题","currentQuestion":"从两张视图到一个可微分的选择题","sourceSlides":[],"knowledgeBlock":null,"formulas":[{"latex":"1/\\tau","definitionContext":"从两张视图到一个可微分的选择题"},{"latex":"p","definitionContext":"从两张视图到一个可微分的选择题"},{"latex":"\\tau=0.5","definitionContext":"从两张视图到一个可微分的选择题"}],"previousBlock":"b-2351d28e0bac3cd3f0f3345c","nextBlock":"b-068167c9c0ee095f5734e8b1","prerequisites":[{"id":"b-cc6657d92ac8ee2789b80c65","url":"#b-cc6657d92ac8ee2789b80c65","role":"section-title"},{"id":"b-36a4e12bf1c1cf36f1a16531","url":"#b-36a4e12bf1c1cf36f1a16531","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-068167c9c0ee095f5734e8b1","type":"p","scope":"document","url":"https://codingai-lec03.pages.dev/course/notebook-blocks/a7a8f9c43bc64c6e8609e87186f7cb77.html#b-068167c9c0ee095f5734e8b1","content":"![同一相似度的温度变化与 MoCo 队列更新](/course/repair/assets/cls06-temperature-queue.svg)","sha256":"1e9d049048183aa6e6e76b4872e6d46739728419873e176868c84cb1e0419b25","chapter":{"id":"course","title":"从两张视图到一个可微分的选择题"},"unit":{"id":"notebook-blocks/a7a8f9c43bc64c6e8609e87186f7cb77","title":"从两张视图到一个可微分的选择题"},"page":{"id":"reading-opening","title":"从两张视图到一个可微分的选择题","number":1},"sectionTitle":"从两张视图到一个可微分的选择题","currentQuestion":"从两张视图到一个可微分的选择题","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-3e9a88c5c0afec4e1aa651d8","nextBlock":"b-9abe50ec4c6816bee42ebb71","prerequisites":[{"id":"b-cc6657d92ac8ee2789b80c65","url":"#b-cc6657d92ac8ee2789b80c65","role":"section-title"},{"id":"b-36a4e12bf1c1cf36f1a16531","url":"#b-36a4e12bf1c1cf36f1a16531","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-9abe50ec4c6816bee42ebb71","type":"p","scope":"document","url":"https://codingai-lec03.pages.dev/course/notebook-blocks/a7a8f9c43bc64c6e8609e87186f7cb77.html#b-9abe50ec4c6816bee42ebb71","content":"上半图按表中真实计算值画出候选概率；下半图是长度 6 的 FIFO 队列，加入两个新 key 后移出两个最老 key。温度控制一次比较，队列控制比较对象从哪里来。","sha256":"5d23cd73ca3e47dcbef8d61f23e2fa0ab3f423d612b6dd4408773577fc82375f","chapter":{"id":"course","title":"从两张视图到一个可微分的选择题"},"unit":{"id":"notebook-blocks/a7a8f9c43bc64c6e8609e87186f7cb77","title":"从两张视图到一个可微分的选择题"},"page":{"id":"reading-opening","title":"从两张视图到一个可微分的选择题","number":1},"sectionTitle":"从两张视图到一个可微分的选择题","currentQuestion":"从两张视图到一个可微分的选择题","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-068167c9c0ee095f5734e8b1","nextBlock":"b-24730bf32aae8b178c77fcb3","prerequisites":[{"id":"b-cc6657d92ac8ee2789b80c65","url":"#b-cc6657d92ac8ee2789b80c65","role":"section-title"},{"id":"b-36a4e12bf1c1cf36f1a16531","url":"#b-36a4e12bf1c1cf36f1a16531","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-24730bf32aae8b178c77fcb3","type":"h2","scope":"document","url":"https://codingai-lec03.pages.dev/course/notebook-blocks/a7a8f9c43bc64c6e8609e87186f7cb77.html#b-24730bf32aae8b178c77fcb3","content":"## 为什么损失作用在 z，迁移却常读 h","sha256":"0c4c6ea20d2153b2e1535a2be1939033ad5c45c42167fba81ef0c782b77d60f1","chapter":{"id":"course","title":"从两张视图到一个可微分的选择题"},"unit":{"id":"notebook-blocks/a7a8f9c43bc64c6e8609e87186f7cb77","title":"从两张视图到一个可微分的选择题"},"page":{"id":"reading-opening","title":"从两张视图到一个可微分的选择题","number":1},"sectionTitle":"从两张视图到一个可微分的选择题","currentQuestion":"从两张视图到一个可微分的选择题","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-9abe50ec4c6816bee42ebb71","nextBlock":"b-be35fb878fa407db71a87d29","prerequisites":[{"id":"b-cc6657d92ac8ee2789b80c65","url":"#b-cc6657d92ac8ee2789b80c65","role":"section-title"},{"id":"b-36a4e12bf1c1cf36f1a16531","url":"#b-36a4e12bf1c1cf36f1a16531","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-be35fb878fa407db71a87d29","type":"p","scope":"document","url":"https://codingai-lec03.pages.dev/course/notebook-blocks/a7a8f9c43bc64c6e8609e87186f7cb77.html#b-be35fb878fa407db71a87d29","content":"假设 $h=(u,c)$，$u$ 编码形状，$c$ 编码颜色。若增强经常改变颜色，两个视图可能有相同 $u$、不同 $c$。直接要求整个 $h$ 相同，会压制颜色差异；增加投影头以后，$g(h)=u$ 这样的映射可以在 $z$ 中满足不变性，而 $h$ 仍可能留下颜色。这是一个**可行性构造**，不是声称真实网络一定学成两根独立坐标。","sha256":"a659d64a58c06d563b62e4fe28734e91d077998a3700743c08340c069019df20","chapter":{"id":"course","title":"从两张视图到一个可微分的选择题"},"unit":{"id":"notebook-blocks/a7a8f9c43bc64c6e8609e87186f7cb77","title":"从两张视图到一个可微分的选择题"},"page":{"id":"reading-opening","title":"从两张视图到一个可微分的选择题","number":1},"sectionTitle":"从两张视图到一个可微分的选择题","currentQuestion":"从两张视图到一个可微分的选择题","sourceSlides":[],"knowledgeBlock":null,"formulas":[{"latex":"h=(u,c)","definitionContext":"从两张视图到一个可微分的选择题"},{"latex":"u","definitionContext":"从两张视图到一个可微分的选择题"},{"latex":"c","definitionContext":"从两张视图到一个可微分的选择题"},{"latex":"u","definitionContext":"从两张视图到一个可微分的选择题"},{"latex":"c","definitionContext":"从两张视图到一个可微分的选择题"},{"latex":"h","definitionContext":"从两张视图到一个可微分的选择题"},{"latex":"g(h)=u","definitionContext":"从两张视图到一个可微分的选择题"},{"latex":"z","definitionContext":"从两张视图到一个可微分的选择题"},{"latex":"h","definitionContext":"从两张视图到一个可微分的选择题"}],"previousBlock":"b-24730bf32aae8b178c77fcb3","nextBlock":"b-91682d0d3d85a1c82e5ecc04","prerequisites":[{"id":"b-24730bf32aae8b178c77fcb3","url":"#b-24730bf32aae8b178c77fcb3","role":"section-title"},{"id":"b-36a4e12bf1c1cf36f1a16531","url":"#b-36a4e12bf1c1cf36f1a16531","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-91682d0d3d85a1c82e5ecc04","type":"p","scope":"document","url":"https://codingai-lec03.pages.dev/course/notebook-blocks/a7a8f9c43bc64c6e8609e87186f7cb77.html#b-91682d0d3d85a1c82e5ecc04","content":"损失计算在 $z$ 不代表编码器没学：链式法则给出 $\\partial L/\\partial\\theta=(\\partial L/\\partial z)(\\partial z/\\partial h)(\\partial h/\\partial\\theta)$。梯度穿过投影头更新编码器。SimCLR 通过实验发现非线性投影头有益，并把投影前表示用于下游；这支持区分目标空间与读出空间，不能推广成“h 永远保留全部被增强抹掉的信息”。图像已经被转成灰度时，该视图根本不含原颜色；若所有训练视图都删掉颜色，表示也没有从这些输入恢复真实颜色的依据。","sha256":"542bbbbbf07478a29b46b11c0e054862781ed72fdbffa19e42b223caf08168a2","chapter":{"id":"course","title":"从两张视图到一个可微分的选择题"},"unit":{"id":"notebook-blocks/a7a8f9c43bc64c6e8609e87186f7cb77","title":"从两张视图到一个可微分的选择题"},"page":{"id":"reading-opening","title":"从两张视图到一个可微分的选择题","number":1},"sectionTitle":"从两张视图到一个可微分的选择题","currentQuestion":"从两张视图到一个可微分的选择题","sourceSlides":[],"knowledgeBlock":null,"formulas":[{"latex":"z","definitionContext":"从两张视图到一个可微分的选择题"},{"latex":"\\partial L/\\partial\\theta=(\\partial L/\\partial z)(\\partial z/\\partial h)(\\partial h/\\partial\\theta)","definitionContext":"从两张视图到一个可微分的选择题"}],"previousBlock":"b-be35fb878fa407db71a87d29","nextBlock":"b-6936c0a1ebe3e8f3e96aee3e","prerequisites":[{"id":"b-24730bf32aae8b178c77fcb3","url":"#b-24730bf32aae8b178c77fcb3","role":"section-title"},{"id":"b-36a4e12bf1c1cf36f1a16531","url":"#b-36a4e12bf1c1cf36f1a16531","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-6936c0a1ebe3e8f3e96aee3e","type":"p","scope":"document","url":"https://codingai-lec03.pages.dev/course/notebook-blocks/a7a8f9c43bc64c6e8609e87186f7cb77.html#b-6936c0a1ebe3e8f3e96aee3e","content":"在 [LAB 06](https://deepnote.com/workspace/ise3309jingao-5ba66b0a-4283-4f5a-ae06-3f68d3201e99/project/38755e4e-23f6-4f5f-8c68-8a2e4114b1c1/notebook/7c4f531113d7473abc174e2e014e1340?utm_source=openai&utm_medium=mcp&utm_campaign=openaimcp&utm_content=7c4f531113d7473abc174e2e014e1340&utm_term=get_notebook) 中，应同时检查颜色/形状任务、增强设置和 $h/z$ 读出。最有解释力的比较是：同一原图划分、相同模型容量与读出预算，只改变颜色增强；再看形状读出和颜色读出是否朝不同方向变化。不能用颜色损失下降代替形状泛化，也不能把随机二维向量当作实际训练出的特征。","sha256":"cdfb9cdcad42224287fc11f2cf2313e29b96100a33841b6a10511d0ba3b8adea","chapter":{"id":"course","title":"从两张视图到一个可微分的选择题"},"unit":{"id":"notebook-blocks/a7a8f9c43bc64c6e8609e87186f7cb77","title":"从两张视图到一个可微分的选择题"},"page":{"id":"reading-opening","title":"从两张视图到一个可微分的选择题","number":1},"sectionTitle":"从两张视图到一个可微分的选择题","currentQuestion":"从两张视图到一个可微分的选择题","sourceSlides":[],"knowledgeBlock":null,"formulas":[{"latex":"h/z","definitionContext":"从两张视图到一个可微分的选择题"}],"previousBlock":"b-91682d0d3d85a1c82e5ecc04","nextBlock":null,"prerequisites":[{"id":"b-24730bf32aae8b178c77fcb3","url":"#b-24730bf32aae8b178c77fcb3","role":"section-title"},{"id":"b-36a4e12bf1c1cf36f1a16531","url":"#b-36a4e12bf1c1cf36f1a16531","role":"problem-context"}],"relatedBlocks":[]}]}