{"version":2,"relationshipURLBase":"https://codingai-lec03.pages.dev/course/notebooks/c0645a4db6e04188ab5b119e87c53bed.html","relationshipResolution":"Resolve relative relationship URLs against relationshipURLBase; complete records and absolute URLs are in blocks.","title":"CLS 06 · 表示的三个检查：目标、信息与新域","url":"https://codingai-lec03.pages.dev/course/notebooks/c0645a4db6e04188ab5b119e87c53bed.html","markdown":"https://codingai-lec03.pages.dev/course/notebooks/c0645a4db6e04188ab5b119e87c53bed.md","pages":[{"id":"reading-opening","title":"CLS 06 · 表示的三个检查：目标、信息与新域","unit":"视图关系怎样成为学习目标，又会牺牲什么？","unitId":"i-view-objective","chapter":"I","number":1},{"id":"reading-block-1861e17c9d7c49f0970e890873be4ff6","title":"教师不一定是更大的模型，也不一定提供真值","unit":"视图关系怎样成为学习目标，又会牺牲什么？","unitId":"i-view-objective","chapter":"I","number":2},{"id":"reading-block-05c08a06e13d4bcb9ad2855d108f9310","title":"先用三个量区分坍塌与无用读出","unit":"视图关系怎样成为学习目标，又会牺牲什么？","unitId":"i-view-objective","chapter":"I","number":3},{"id":"reading-block-0d856b8e3bd442f9bd30fd3a2b3a9d60","title":"分布变了，源域最方便的线索可能最危险","unit":"视图关系怎样成为学习目标，又会牺牲什么？","unitId":"i-view-objective","chapter":"I","number":4}],"aliases":[{"oldId":"b-f799dd9ee05dd1af26aca0cf","targetId":"b-f799dd9ee05dd1af26aca0cf","page":"reading-opening","operation":"move","beforeSHA256":"cbc29af1a12f4af0f029d874fd4a12bc8e9062d888d9db399a00c348c196ef30","afterSHA256":"cbc29af1a12f4af0f029d874fd4a12bc8e9062d888d9db399a00c348c196ef30","reason":"Same source body in authored semantic page"},{"oldId":"b-21060f5df9d972c6533f2bd0","targetId":"b-21060f5df9d972c6533f2bd0","page":"reading-opening","operation":"move","beforeSHA256":"e91f24e4e4ca40e2099a0d81d30db6568f4bf2074b499eeee6e3959617bd97ac","afterSHA256":"e91f24e4e4ca40e2099a0d81d30db6568f4bf2074b499eeee6e3959617bd97ac","reason":"Same source body in authored semantic page"},{"oldId":"b-95b6abeafd3a37576deed778","targetId":"b-95b6abeafd3a37576deed778","page":"reading-opening","operation":"move","beforeSHA256":"44af65f60193d6766e8cc6019f133e85a56a9751792313a8cd2e06a0256ec27f","afterSHA256":"44af65f60193d6766e8cc6019f133e85a56a9751792313a8cd2e06a0256ec27f","reason":"Same source body in authored semantic page"},{"oldId":"b-69f2cb67d862cecfa4dd5a01","targetId":"b-69f2cb67d862cecfa4dd5a01","page":"reading-opening","operation":"move","beforeSHA256":"f22152fd55dcdb0090b5653d37145694fd19388d31e10226549464fe8b483578","afterSHA256":"f22152fd55dcdb0090b5653d37145694fd19388d31e10226549464fe8b483578","reason":"Same source body in authored semantic page"},{"oldId":"b-dd468cc7f391ed72b8e5cf90","targetId":"b-dd468cc7f391ed72b8e5cf90","page":"reading-block-1861e17c9d7c49f0970e890873be4ff6","operation":"move","beforeSHA256":"efc07be0d238c97b2a0f2a3e1e141f27691e7d1aadb5b0bc2662d7d40218327f","afterSHA256":"efc07be0d238c97b2a0f2a3e1e141f27691e7d1aadb5b0bc2662d7d40218327f","reason":"Same source body in authored semantic page"},{"oldId":"b-327fe729d7d5c7af35b80a13","targetId":"b-327fe729d7d5c7af35b80a13","page":"reading-block-1861e17c9d7c49f0970e890873be4ff6","operation":"move","beforeSHA256":"7a09424a9fcff03809af57ece9eea1d6c350b11893d6409472958f4ef32264b2","afterSHA256":"7a09424a9fcff03809af57ece9eea1d6c350b11893d6409472958f4ef32264b2","reason":"Same source body in authored semantic page"},{"oldId":"b-a2fb5f81c816fd27162e36e7","targetId":"b-a2fb5f81c816fd27162e36e7","page":"reading-block-1861e17c9d7c49f0970e890873be4ff6","operation":"move","beforeSHA256":"e15af766bfdd95248a72eee449be07b21825db5fb82400539f177dbfbcf75e04","afterSHA256":"e15af766bfdd95248a72eee449be07b21825db5fb82400539f177dbfbcf75e04","reason":"Same source body in authored semantic page"},{"oldId":"b-2029ea76cd504ed44c4e790f","targetId":"b-2029ea76cd504ed44c4e790f","page":"reading-block-1861e17c9d7c49f0970e890873be4ff6","operation":"move","beforeSHA256":"a4194bffb03d98e3ab6ee0e192440540c4b75b3bbddb7d347c41aed9223cd2d7","afterSHA256":"a4194bffb03d98e3ab6ee0e192440540c4b75b3bbddb7d347c41aed9223cd2d7","reason":"Same source body in authored semantic page"},{"oldId":"b-63738d1d58a89e51cfd712ee","targetId":"b-63738d1d58a89e51cfd712ee","page":"reading-block-1861e17c9d7c49f0970e890873be4ff6","operation":"move","beforeSHA256":"0e85b26615f1bb8ce7feef1df33b086ae214edf48843560f3591ee71d79f0e2e","afterSHA256":"0e85b26615f1bb8ce7feef1df33b086ae214edf48843560f3591ee71d79f0e2e","reason":"Same source body in authored semantic page"},{"oldId":"b-4499f5a4d20b1ae9639b1693","targetId":"b-4499f5a4d20b1ae9639b1693","page":"reading-block-1861e17c9d7c49f0970e890873be4ff6","operation":"move","beforeSHA256":"2ef2ec8a8fd967adffc562707501137a4fcc967a2af3e53e4d167b0e0f1aff51","afterSHA256":"2ef2ec8a8fd967adffc562707501137a4fcc967a2af3e53e4d167b0e0f1aff51","reason":"Same source body in authored semantic page"},{"oldId":"b-eaf25dac67d136743fe9e50b","targetId":"b-eaf25dac67d136743fe9e50b","page":"reading-block-1861e17c9d7c49f0970e890873be4ff6","operation":"move","beforeSHA256":"4730b4cad436cd4b274c52e198e7548d06b30738b1d878fb35771460fd51ec8f","afterSHA256":"4730b4cad436cd4b274c52e198e7548d06b30738b1d878fb35771460fd51ec8f","reason":"Same source body in authored semantic page"},{"oldId":"b-e69ca1882f24fcbee8cda6c3","targetId":"b-e69ca1882f24fcbee8cda6c3","page":"reading-block-1861e17c9d7c49f0970e890873be4ff6","operation":"move","beforeSHA256":"18d41b1e5728c89da6ca0c68d3702e0dd7c3d817a177e150d1c544abfb081458","afterSHA256":"18d41b1e5728c89da6ca0c68d3702e0dd7c3d817a177e150d1c544abfb081458","reason":"Same source body in authored semantic page"},{"oldId":"b-2b528ea2077457439efbcb3f","targetId":"b-2b528ea2077457439efbcb3f","page":"reading-block-1861e17c9d7c49f0970e890873be4ff6","operation":"move","beforeSHA256":"9daa40692cb1e71c17ff59655c00da11e6e36068fd5ca5549da5016ecb2f031e","afterSHA256":"9daa40692cb1e71c17ff59655c00da11e6e36068fd5ca5549da5016ecb2f031e","reason":"Same source body in authored semantic page"},{"oldId":"b-fb413d3589c141432f2b49c1","targetId":"b-fb413d3589c141432f2b49c1","page":"reading-block-1861e17c9d7c49f0970e890873be4ff6","operation":"move","beforeSHA256":"60c0021cf3e308ab09b33e2eaa8dcc837399ca1d7ea911e4c5988bde77692bd7","afterSHA256":"60c0021cf3e308ab09b33e2eaa8dcc837399ca1d7ea911e4c5988bde77692bd7","reason":"Same source body in authored semantic page"},{"oldId":"b-c5e3b03dbe391904362b875d","targetId":"b-c5e3b03dbe391904362b875d","page":"reading-block-1861e17c9d7c49f0970e890873be4ff6","operation":"move","beforeSHA256":"75a4add903cc4a1e6fb297730fba5165423f8e20dbba8243cadb62f713b8be20","afterSHA256":"75a4add903cc4a1e6fb297730fba5165423f8e20dbba8243cadb62f713b8be20","reason":"Same source body in authored semantic page"},{"oldId":"b-3a00eb35cf5acf2e4c59cd44","targetId":"b-3a00eb35cf5acf2e4c59cd44","page":"reading-block-1861e17c9d7c49f0970e890873be4ff6","operation":"move","beforeSHA256":"792d216843f4f01968151809602fa8c84963defa7cf462797140948c3f044a6b","afterSHA256":"792d216843f4f01968151809602fa8c84963defa7cf462797140948c3f044a6b","reason":"Same source body in authored semantic page"},{"oldId":"b-23aeb83178fa4aa532191170","targetId":"b-23aeb83178fa4aa532191170","page":"reading-block-1861e17c9d7c49f0970e890873be4ff6","operation":"rewrite","beforeSHA256":"c21902cabd20ed4ed1f16ebb7e6e27c94c38ea438454af6ee6ffb747aabec209","afterSHA256":"5bea10eb69b97fe9ff1f3306fc82766b9958278943003d0ce821c7862c98df24","reason":"Same source identity with figure semantics or reading context enriched"},{"oldId":"b-eb294f85f5d52b98de0ce3f6","targetId":"b-eb294f85f5d52b98de0ce3f6","page":"reading-block-1861e17c9d7c49f0970e890873be4ff6","operation":"move","beforeSHA256":"ad1c9cdc8c931315cefa1f74029155573a31c600a5d190e736b1616f4743d31a","afterSHA256":"ad1c9cdc8c931315cefa1f74029155573a31c600a5d190e736b1616f4743d31a","reason":"Same source body in authored semantic page"},{"oldId":"b-cd1df5b97055d6e0f9dd91ee","targetId":"b-cd1df5b97055d6e0f9dd91ee","page":"reading-block-05c08a06e13d4bcb9ad2855d108f9310","operation":"move","beforeSHA256":"7b046569644a82f65ad052205227f16c1b004172e34365cbe4de73c52681b420","afterSHA256":"7b046569644a82f65ad052205227f16c1b004172e34365cbe4de73c52681b420","reason":"Same source body in authored semantic page"},{"oldId":"b-1051bf024f760b86e31743c8","targetId":"b-1051bf024f760b86e31743c8","page":"reading-block-05c08a06e13d4bcb9ad2855d108f9310","operation":"move","beforeSHA256":"3c4a62216e8f3c52833f83e5a8ceb945b5bdb027d0b6deabca56ad1fba8c71e9","afterSHA256":"3c4a62216e8f3c52833f83e5a8ceb945b5bdb027d0b6deabca56ad1fba8c71e9","reason":"Same source body in authored semantic page"},{"oldId":"b-02820f378c7ad319163233f4","targetId":"b-02820f378c7ad319163233f4","page":"reading-block-05c08a06e13d4bcb9ad2855d108f9310","operation":"move","beforeSHA256":"5cfb5233b27d8709e56ed0dc88b6b62514c0f383716469d139b170d4beb2f439","afterSHA256":"5cfb5233b27d8709e56ed0dc88b6b62514c0f383716469d139b170d4beb2f439","reason":"Same source body in authored semantic page"},{"oldId":"b-74dc730edd62ab4322618a76","targetId":"b-74dc730edd62ab4322618a76","page":"reading-block-05c08a06e13d4bcb9ad2855d108f9310","operation":"move","beforeSHA256":"3dea9621641a48772801e1ca7f637ce261694961d4996c34799d658eb0295c99","afterSHA256":"3dea9621641a48772801e1ca7f637ce261694961d4996c34799d658eb0295c99","reason":"Same source body in authored semantic page"},{"oldId":"b-2586af661eb44d4b4ae610ad","targetId":"b-2586af661eb44d4b4ae610ad","page":"reading-block-05c08a06e13d4bcb9ad2855d108f9310","operation":"move","beforeSHA256":"b8b70ff82c0bcd32ddaf4e44dea0f80f7c67697b267456c27d17cf9879c89401","afterSHA256":"b8b70ff82c0bcd32ddaf4e44dea0f80f7c67697b267456c27d17cf9879c89401","reason":"Same source body in authored semantic page"},{"oldId":"b-03d863e5189e6836edf94b2a","targetId":"b-03d863e5189e6836edf94b2a","page":"reading-block-05c08a06e13d4bcb9ad2855d108f9310","operation":"move","beforeSHA256":"30c4d3d14ad50393c706ffbf805526ef1d5433c63b4d438f68c0a798cc524ae6","afterSHA256":"30c4d3d14ad50393c706ffbf805526ef1d5433c63b4d438f68c0a798cc524ae6","reason":"Same source body in authored semantic page"},{"oldId":"b-d16040819ba81a3bdfe0bf80","targetId":"b-d16040819ba81a3bdfe0bf80","page":"reading-block-05c08a06e13d4bcb9ad2855d108f9310","operation":"move","beforeSHA256":"f297f61db6a828793849fb1c0a83af36c1b9e862d1024cebda5de5f7fcdee6ff","afterSHA256":"f297f61db6a828793849fb1c0a83af36c1b9e862d1024cebda5de5f7fcdee6ff","reason":"Same source body in authored semantic page"},{"oldId":"b-d0db3b6e0edd6a6be0672c22","targetId":"b-d0db3b6e0edd6a6be0672c22","page":"reading-block-05c08a06e13d4bcb9ad2855d108f9310","operation":"move","beforeSHA256":"0fcfc28644b12ce434e17f07aa61a599852cad27f630da57f20efd4c7ad4037d","afterSHA256":"0fcfc28644b12ce434e17f07aa61a599852cad27f630da57f20efd4c7ad4037d","reason":"Same source body in authored semantic page"},{"oldId":"b-26330905fc670d466b3f9e7c","targetId":"b-26330905fc670d466b3f9e7c","page":"reading-block-05c08a06e13d4bcb9ad2855d108f9310","operation":"move","beforeSHA256":"aa3bbe0b7cc6c328d10bb19ad23b01d3e3c11498cbcd139289bbbc11c3e6c9a1","afterSHA256":"aa3bbe0b7cc6c328d10bb19ad23b01d3e3c11498cbcd139289bbbc11c3e6c9a1","reason":"Same source body in authored semantic page"},{"oldId":"b-c6ee3368384109f6a605e6ef","targetId":"b-c6ee3368384109f6a605e6ef","page":"reading-block-0d856b8e3bd442f9bd30fd3a2b3a9d60","operation":"move","beforeSHA256":"718136f0278e8a0ff5a30f1c38d5890c1f388a5bd212c24222a2e8e229889387","afterSHA256":"718136f0278e8a0ff5a30f1c38d5890c1f388a5bd212c24222a2e8e229889387","reason":"Same source body in authored semantic page"},{"oldId":"b-28c611885c13f84f89e1ae3e","targetId":"b-28c611885c13f84f89e1ae3e","page":"reading-block-0d856b8e3bd442f9bd30fd3a2b3a9d60","operation":"move","beforeSHA256":"60c3e09851226bcb5159f1fda0394bde192080355322a0673534905c9090a271","afterSHA256":"60c3e09851226bcb5159f1fda0394bde192080355322a0673534905c9090a271","reason":"Same source body in authored semantic page"},{"oldId":"b-a1d3a29c5b5b7f81185c49ff","targetId":"b-a1d3a29c5b5b7f81185c49ff","page":"reading-block-0d856b8e3bd442f9bd30fd3a2b3a9d60","operation":"move","beforeSHA256":"b655be870fde80e77afd34b140e3a1a1938336d19183cb649e78ee4e026567d1","afterSHA256":"b655be870fde80e77afd34b140e3a1a1938336d19183cb649e78ee4e026567d1","reason":"Same source body in authored semantic page"},{"oldId":"b-8cc7eba8293e6e1d91eaf1e0","targetId":"b-8cc7eba8293e6e1d91eaf1e0","page":"reading-block-0d856b8e3bd442f9bd30fd3a2b3a9d60","operation":"move","beforeSHA256":"d622785e68c1be46b0cab51204b3da00c5c09a4111af63a00a58898ce3ad502c","afterSHA256":"d622785e68c1be46b0cab51204b3da00c5c09a4111af63a00a58898ce3ad502c","reason":"Same source body in authored semantic page"},{"oldId":"b-6d9778373e0abd007e832054","targetId":"b-6d9778373e0abd007e832054","page":"reading-block-0d856b8e3bd442f9bd30fd3a2b3a9d60","operation":"move","beforeSHA256":"12bcd5d53dcdcdd3a25813b2723f5fe32ed93be252405fcf1d9df668daa7d8a8","afterSHA256":"12bcd5d53dcdcdd3a25813b2723f5fe32ed93be252405fcf1d9df668daa7d8a8","reason":"Same source body in authored semantic page"},{"oldId":"block-1861e17c9d7c49f0970e890873be4ff6--教师不一定是更大的模型也不一定提供真值","targetId":"b-2b528ea2077457439efbcb3f","operation":"retain-source-identity"},{"oldId":"block-05c08a06e13d4bcb9ad2855d108f9310--先用三个量区分坍塌与无用读出","targetId":"b-cd1df5b97055d6e0f9dd91ee","operation":"retain-source-identity"},{"oldId":"block-05c08a06e13d4bcb9ad2855d108f9310--数据处理不等式究竟限制哪一条链","targetId":"b-d16040819ba81a3bdfe0bf80","operation":"retain-source-identity"},{"oldId":"block-0d856b8e3bd442f9bd30fd3a2b3a9d60--分布变了源域最方便的线索可能最危险","targetId":"b-c6ee3368384109f6a605e6ef","operation":"retain-source-identity"}],"blocks":[{"id":"b-f799dd9ee05dd1af26aca0cf","type":"h1","scope":"document","url":"https://codingai-lec03.pages.dev/course/notebooks/c0645a4db6e04188ab5b119e87c53bed.html#b-f799dd9ee05dd1af26aca0cf","content":"# CLS 06 · 表示的三个检查：目标、信息与新域","sha256":"cbc29af1a12f4af0f029d874fd4a12bc8e9062d888d9db399a00c348c196ef30","chapter":{"id":"I","title":"学习信号与表示"},"unit":{"id":"i-view-objective","title":"视图关系怎样成为学习目标，又会牺牲什么？"},"page":{"id":"reading-opening","title":"CLS 06 · 表示的三个检查：目标、信息与新域","number":1},"sectionTitle":"CLS 06 · 表示的三个检查：目标、信息与新域","currentQuestion":"CLS 06 · 表示的三个检查：目标、信息与新域","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":null,"nextBlock":"b-21060f5df9d972c6533f2bd0","prerequisites":[{"id":"b-21060f5df9d972c6533f2bd0","url":"#b-21060f5df9d972c6533f2bd0","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-21060f5df9d972c6533f2bd0","type":"p","scope":"document","url":"https://codingai-lec03.pages.dev/course/notebooks/c0645a4db6e04188ab5b119e87c53bed.html#b-21060f5df9d972c6533f2bd0","content":"按问题继续","sha256":"e91f24e4e4ca40e2099a0d81d30db6568f4bf2074b499eeee6e3959617bd97ac","chapter":{"id":"I","title":"学习信号与表示"},"unit":{"id":"i-view-objective","title":"视图关系怎样成为学习目标，又会牺牲什么？"},"page":{"id":"reading-opening","title":"CLS 06 · 表示的三个检查：目标、信息与新域","number":1},"sectionTitle":"CLS 06 · 表示的三个检查：目标、信息与新域","currentQuestion":"CLS 06 · 表示的三个检查：目标、信息与新域","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-f799dd9ee05dd1af26aca0cf","nextBlock":"b-95b6abeafd3a37576deed778","prerequisites":[{"id":"b-f799dd9ee05dd1af26aca0cf","url":"#b-f799dd9ee05dd1af26aca0cf","role":"section-title"}],"relatedBlocks":[]},{"id":"b-95b6abeafd3a37576deed778","type":"p","scope":"document","url":"https://codingai-lec03.pages.dev/course/notebooks/c0645a4db6e04188ab5b119e87c53bed.html#b-95b6abeafd3a37576deed778","content":"教师软目标、坍塌、任务信息与域变化的三个不同问题","sha256":"44af65f60193d6766e8cc6019f133e85a56a9751792313a8cd2e06a0256ec27f","chapter":{"id":"I","title":"学习信号与表示"},"unit":{"id":"i-view-objective","title":"视图关系怎样成为学习目标，又会牺牲什么？"},"page":{"id":"reading-opening","title":"CLS 06 · 表示的三个检查：目标、信息与新域","number":1},"sectionTitle":"CLS 06 · 表示的三个检查：目标、信息与新域","currentQuestion":"CLS 06 · 表示的三个检查：目标、信息与新域","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-21060f5df9d972c6533f2bd0","nextBlock":"b-69f2cb67d862cecfa4dd5a01","prerequisites":[{"id":"b-f799dd9ee05dd1af26aca0cf","url":"#b-f799dd9ee05dd1af26aca0cf","role":"section-title"},{"id":"b-21060f5df9d972c6533f2bd0","url":"#b-21060f5df9d972c6533f2bd0","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-69f2cb67d862cecfa4dd5a01","type":"p","scope":"document","url":"https://codingai-lec03.pages.dev/course/notebooks/c0645a4db6e04188ab5b119e87c53bed.html#b-69f2cb67d862cecfa4dd5a01","content":"可先阅读已有结果，再决定是否运行。重算时请核对各节记录的数据、参数、环境与日期。","sha256":"f22152fd55dcdb0090b5653d37145694fd19388d31e10226549464fe8b483578","chapter":{"id":"I","title":"学习信号与表示"},"unit":{"id":"i-view-objective","title":"视图关系怎样成为学习目标，又会牺牲什么？"},"page":{"id":"reading-opening","title":"CLS 06 · 表示的三个检查：目标、信息与新域","number":1},"sectionTitle":"CLS 06 · 表示的三个检查：目标、信息与新域","currentQuestion":"CLS 06 · 表示的三个检查：目标、信息与新域","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-95b6abeafd3a37576deed778","nextBlock":null,"prerequisites":[{"id":"b-f799dd9ee05dd1af26aca0cf","url":"#b-f799dd9ee05dd1af26aca0cf","role":"section-title"},{"id":"b-21060f5df9d972c6533f2bd0","url":"#b-21060f5df9d972c6533f2bd0","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-dd468cc7f391ed72b8e5cf90","type":"p","scope":"block-1861e17c9d7c49f0970e890873be4ff6","url":"https://codingai-lec03.pages.dev/course/notebooks/c0645a4db6e04188ab5b119e87c53bed.html#b-dd468cc7f391ed72b8e5cf90","content":"**学习路径：** [表示怎样迁移到新任务](/course/study/readout-transfer.html)","sha256":"efc07be0d238c97b2a0f2a3e1e141f27691e7d1aadb5b0bc2662d7d40218327f","chapter":{"id":"I","title":"学习信号与表示"},"unit":{"id":"i-view-objective","title":"视图关系怎样成为学习目标，又会牺牲什么？"},"page":{"id":"reading-block-1861e17c9d7c49f0970e890873be4ff6","title":"教师不一定是更大的模型，也不一定提供真值","number":2},"sectionTitle":"教师不一定是更大的模型，也不一定提供真值","currentQuestion":"教师不一定是更大的模型，也不一定提供真值","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":null,"nextBlock":"b-327fe729d7d5c7af35b80a13","prerequisites":[],"relatedBlocks":[]},{"id":"b-327fe729d7d5c7af35b80a13","type":"p","scope":"block-1861e17c9d7c49f0970e890873be4ff6","url":"https://codingai-lec03.pages.dev/course/notebooks/c0645a4db6e04188ab5b119e87c53bed.html#b-327fe729d7d5c7af35b80a13","content":"从[P23评价协议](/course/pages/l3-r2-p23.html)或[P24新任务](/course/pages/l3-r2-p24.html)带一个问题进入，不必顺序读完三条分支。","sha256":"7a09424a9fcff03809af57ece9eea1d6c350b11893d6409472958f4ef32264b2","chapter":{"id":"I","title":"学习信号与表示"},"unit":{"id":"i-view-objective","title":"视图关系怎样成为学习目标，又会牺牲什么？"},"page":{"id":"reading-block-1861e17c9d7c49f0970e890873be4ff6","title":"教师不一定是更大的模型，也不一定提供真值","number":2},"sectionTitle":"教师不一定是更大的模型，也不一定提供真值","currentQuestion":"教师不一定是更大的模型，也不一定提供真值","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-dd468cc7f391ed72b8e5cf90","nextBlock":"b-a2fb5f81c816fd27162e36e7","prerequisites":[{"id":"b-dd468cc7f391ed72b8e5cf90","url":"#b-dd468cc7f391ed72b8e5cf90","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-a2fb5f81c816fd27162e36e7","type":"ul","scope":"block-1861e17c9d7c49f0970e890873be4ff6","url":"https://codingai-lec03.pages.dev/course/notebooks/c0645a4db6e04188ab5b119e87c53bed.html#b-a2fb5f81c816fd27162e36e7","content":"- **目标是否允许无用的解？** 看常量表示为何得到零一致性损失，再用[LAB 17](https://deepnote.com/project/38755e4e-23f6-4f5f-8c68-8a2e4114b1c1/notebook/268a4ab11b464c64b065f9fc12b19cc7)区分固定教师拟合与独立EMA计算。\n\n- **线性读不出是否等于信息丢失？** 看[XOR四点图](/course/blocks/information-and-readable-form.html)，再用[LAB 18](https://deepnote.com/project/38755e4e-23f6-4f5f-8c68-8a2e4114b1c1/notebook/33066318f5884adc8417068df0f40431)比较标签信息、身份信息和读出形式。\n\n- **换环境失效是否就是负迁移？** 看[环境变化与负迁移](/course/blocks/environment-shift-versus-negative-transfer.html)，再用[LAB 19](https://deepnote.com/project/38755e4e-23f6-4f5f-8c68-8a2e4114b1c1/notebook/c20ef6a65d184331b1b52bf351077504)分别计算适配收益与相对scratch收益。","sha256":"e15af766bfdd95248a72eee449be07b21825db5fb82400539f177dbfbcf75e04","chapter":{"id":"I","title":"学习信号与表示"},"unit":{"id":"i-view-objective","title":"视图关系怎样成为学习目标，又会牺牲什么？"},"page":{"id":"reading-block-1861e17c9d7c49f0970e890873be4ff6","title":"教师不一定是更大的模型，也不一定提供真值","number":2},"sectionTitle":"教师不一定是更大的模型，也不一定提供真值","currentQuestion":"教师不一定是更大的模型，也不一定提供真值","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-327fe729d7d5c7af35b80a13","nextBlock":"b-2029ea76cd504ed44c4e790f","prerequisites":[{"id":"b-dd468cc7f391ed72b8e5cf90","url":"#b-dd468cc7f391ed72b8e5cf90","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-2029ea76cd504ed44c4e790f","type":"li","scope":"block-1861e17c9d7c49f0970e890873be4ff6","url":"https://codingai-lec03.pages.dev/course/notebooks/c0645a4db6e04188ab5b119e87c53bed.html#b-2029ea76cd504ed44c4e790f","content":"- **目标是否允许无用的解？** 看常量表示为何得到零一致性损失，再用[LAB 17](https://deepnote.com/project/38755e4e-23f6-4f5f-8c68-8a2e4114b1c1/notebook/268a4ab11b464c64b065f9fc12b19cc7)区分固定教师拟合与独立EMA计算。","sha256":"a4194bffb03d98e3ab6ee0e192440540c4b75b3bbddb7d347c41aed9223cd2d7","chapter":{"id":"I","title":"学习信号与表示"},"unit":{"id":"i-view-objective","title":"视图关系怎样成为学习目标，又会牺牲什么？"},"page":{"id":"reading-block-1861e17c9d7c49f0970e890873be4ff6","title":"教师不一定是更大的模型，也不一定提供真值","number":2},"sectionTitle":"教师不一定是更大的模型，也不一定提供真值","currentQuestion":"教师不一定是更大的模型，也不一定提供真值","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-a2fb5f81c816fd27162e36e7","nextBlock":"b-63738d1d58a89e51cfd712ee","prerequisites":[{"id":"b-dd468cc7f391ed72b8e5cf90","url":"#b-dd468cc7f391ed72b8e5cf90","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-63738d1d58a89e51cfd712ee","type":"li","scope":"block-1861e17c9d7c49f0970e890873be4ff6","url":"https://codingai-lec03.pages.dev/course/notebooks/c0645a4db6e04188ab5b119e87c53bed.html#b-63738d1d58a89e51cfd712ee","content":"- **线性读不出是否等于信息丢失？** 看[XOR四点图](/course/blocks/information-and-readable-form.html)，再用[LAB 18](https://deepnote.com/project/38755e4e-23f6-4f5f-8c68-8a2e4114b1c1/notebook/33066318f5884adc8417068df0f40431)比较标签信息、身份信息和读出形式。","sha256":"0e85b26615f1bb8ce7feef1df33b086ae214edf48843560f3591ee71d79f0e2e","chapter":{"id":"I","title":"学习信号与表示"},"unit":{"id":"i-view-objective","title":"视图关系怎样成为学习目标，又会牺牲什么？"},"page":{"id":"reading-block-1861e17c9d7c49f0970e890873be4ff6","title":"教师不一定是更大的模型，也不一定提供真值","number":2},"sectionTitle":"教师不一定是更大的模型，也不一定提供真值","currentQuestion":"教师不一定是更大的模型，也不一定提供真值","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-2029ea76cd504ed44c4e790f","nextBlock":"b-4499f5a4d20b1ae9639b1693","prerequisites":[{"id":"b-dd468cc7f391ed72b8e5cf90","url":"#b-dd468cc7f391ed72b8e5cf90","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-4499f5a4d20b1ae9639b1693","type":"li","scope":"block-1861e17c9d7c49f0970e890873be4ff6","url":"https://codingai-lec03.pages.dev/course/notebooks/c0645a4db6e04188ab5b119e87c53bed.html#b-4499f5a4d20b1ae9639b1693","content":"- **换环境失效是否就是负迁移？** 看[环境变化与负迁移](/course/blocks/environment-shift-versus-negative-transfer.html)，再用[LAB 19](https://deepnote.com/project/38755e4e-23f6-4f5f-8c68-8a2e4114b1c1/notebook/c20ef6a65d184331b1b52bf351077504)分别计算适配收益与相对scratch收益。","sha256":"2ef2ec8a8fd967adffc562707501137a4fcc967a2af3e53e4d167b0e0f1aff51","chapter":{"id":"I","title":"学习信号与表示"},"unit":{"id":"i-view-objective","title":"视图关系怎样成为学习目标，又会牺牲什么？"},"page":{"id":"reading-block-1861e17c9d7c49f0970e890873be4ff6","title":"教师不一定是更大的模型，也不一定提供真值","number":2},"sectionTitle":"教师不一定是更大的模型，也不一定提供真值","currentQuestion":"教师不一定是更大的模型，也不一定提供真值","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-63738d1d58a89e51cfd712ee","nextBlock":"b-eaf25dac67d136743fe9e50b","prerequisites":[{"id":"b-dd468cc7f391ed72b8e5cf90","url":"#b-dd468cc7f391ed72b8e5cf90","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-eaf25dac67d136743fe9e50b","type":"p","scope":"block-1861e17c9d7c49f0970e890873be4ff6","url":"https://codingai-lec03.pages.dev/course/notebooks/c0645a4db6e04188ab5b119e87c53bed.html#b-eaf25dac67d136743fe9e50b","content":"留下一个可检验判断即可：指出你比较的是哪两个对象、固定了什么条件、结果排除了哪种解释。若要用真实图像检查冻结与微调，回到[LAB 01](https://deepnote.com/project/38755e4e-23f6-4f5f-8c68-8a2e4114b1c1/notebook/4dd8605ed9904820919ea9e13d9e7d7c)/[LAB 02](https://deepnote.com/project/38755e4e-23f6-4f5f-8c68-8a2e4114b1c1/notebook/4bd450d54870432db91c80a4df563b8e)。","sha256":"4730b4cad436cd4b274c52e198e7548d06b30738b1d878fb35771460fd51ec8f","chapter":{"id":"I","title":"学习信号与表示"},"unit":{"id":"i-view-objective","title":"视图关系怎样成为学习目标，又会牺牲什么？"},"page":{"id":"reading-block-1861e17c9d7c49f0970e890873be4ff6","title":"教师不一定是更大的模型，也不一定提供真值","number":2},"sectionTitle":"教师不一定是更大的模型，也不一定提供真值","currentQuestion":"教师不一定是更大的模型，也不一定提供真值","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-4499f5a4d20b1ae9639b1693","nextBlock":"b-e69ca1882f24fcbee8cda6c3","prerequisites":[{"id":"b-dd468cc7f391ed72b8e5cf90","url":"#b-dd468cc7f391ed72b8e5cf90","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-e69ca1882f24fcbee8cda6c3","type":"p","scope":"block-1861e17c9d7c49f0970e890873be4ff6","url":"https://codingai-lec03.pages.dev/course/notebooks/c0645a4db6e04188ab5b119e87c53bed.html#b-e69ca1882f24fcbee8cda6c3","content":"对比学习用负例阻止所有样本挤成一点。如果去掉负例，只让同图两视图接近，常量表示 $f(x)=c$ 对任意输入都能达到零距离。这个解完全满足“彼此一致”，却不能区分杯子和书。于是问题从“怎样产生训练目标”推进为：目标本身是否排除了我们不想要的解？本册把教师学习、表示信息和域变化连起来，是因为三者都要求分清**模型容易满足的条件**和**我们真正需要的能力**。","sha256":"18d41b1e5728c89da6ca0c68d3702e0dd7c3d817a177e150d1c544abfb081458","chapter":{"id":"I","title":"学习信号与表示"},"unit":{"id":"i-view-objective","title":"视图关系怎样成为学习目标，又会牺牲什么？"},"page":{"id":"reading-block-1861e17c9d7c49f0970e890873be4ff6","title":"教师不一定是更大的模型，也不一定提供真值","number":2},"sectionTitle":"教师不一定是更大的模型，也不一定提供真值","currentQuestion":"教师不一定是更大的模型，也不一定提供真值","sourceSlides":[],"knowledgeBlock":null,"formulas":[{"latex":"f(x)=c","definitionContext":"教师不一定是更大的模型，也不一定提供真值"}],"previousBlock":"b-eaf25dac67d136743fe9e50b","nextBlock":"b-2b528ea2077457439efbcb3f","prerequisites":[{"id":"b-dd468cc7f391ed72b8e5cf90","url":"#b-dd468cc7f391ed72b8e5cf90","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-2b528ea2077457439efbcb3f","type":"h2","scope":"block-1861e17c9d7c49f0970e890873be4ff6","url":"https://codingai-lec03.pages.dev/course/notebooks/c0645a4db6e04188ab5b119e87c53bed.html#b-2b528ea2077457439efbcb3f","content":"## 教师不一定是更大的模型，也不一定提供真值","sha256":"9daa40692cb1e71c17ff59655c00da11e6e36068fd5ca5549da5016ecb2f031e","chapter":{"id":"I","title":"学习信号与表示"},"unit":{"id":"i-view-objective","title":"视图关系怎样成为学习目标，又会牺牲什么？"},"page":{"id":"reading-block-1861e17c9d7c49f0970e890873be4ff6","title":"教师不一定是更大的模型，也不一定提供真值","number":2},"sectionTitle":"教师不一定是更大的模型，也不一定提供真值","currentQuestion":"教师不一定是更大的模型，也不一定提供真值","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-e69ca1882f24fcbee8cda6c3","nextBlock":"b-fb413d3589c141432f2b49c1","prerequisites":[{"id":"b-dd468cc7f391ed72b8e5cf90","url":"#b-dd468cc7f391ed72b8e5cf90","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-fb413d3589c141432f2b49c1","type":"p","scope":"block-1861e17c9d7c49f0970e890873be4ff6","url":"https://codingai-lec03.pages.dev/course/notebooks/c0645a4db6e04188ab5b119e87c53bed.html#b-fb413d3589c141432f2b49c1","content":"传统知识蒸馏用训练好的教师给学生提供输出分布。教师 logits $a=(4,2,0)$ 对应三个类别，温度 $T$ 下的软目标为 $q_i=\\exp(a_i/T)/\\sum_j\\exp(a_j/T)$。$T=1$ 得到约 $(0.8668,0.1173,0.0159)$，$T=2$ 得到 $(0.6652,0.2447,0.0900)$。硬标签只指出第一类；软目标还表达第二类比第三类更接近教师判断。学生可用交叉熵 $-\\sum_iq_i\\log p_i$ 或等价的、忽略教师熵常数后的 KL 项拟合。","sha256":"60c0021cf3e308ab09b33e2eaa8dcc837399ca1d7ea911e4c5988bde77692bd7","chapter":{"id":"I","title":"学习信号与表示"},"unit":{"id":"i-view-objective","title":"视图关系怎样成为学习目标，又会牺牲什么？"},"page":{"id":"reading-block-1861e17c9d7c49f0970e890873be4ff6","title":"教师不一定是更大的模型，也不一定提供真值","number":2},"sectionTitle":"教师不一定是更大的模型，也不一定提供真值","currentQuestion":"教师不一定是更大的模型，也不一定提供真值","sourceSlides":[],"knowledgeBlock":null,"formulas":[{"latex":"a=(4,2,0)","definitionContext":"教师不一定是更大的模型，也不一定提供真值"},{"latex":"T","definitionContext":"教师不一定是更大的模型，也不一定提供真值"},{"latex":"q_i=\\exp(a_i/T)/\\sum_j\\exp(a_j/T)","definitionContext":"教师不一定是更大的模型，也不一定提供真值"},{"latex":"T=1","definitionContext":"教师不一定是更大的模型，也不一定提供真值"},{"latex":"(0.8668,0.1173,0.0159)","definitionContext":"教师不一定是更大的模型，也不一定提供真值"},{"latex":"T=2","definitionContext":"教师不一定是更大的模型，也不一定提供真值"},{"latex":"(0.6652,0.2447,0.0900)","definitionContext":"教师不一定是更大的模型，也不一定提供真值"},{"latex":"-\\sum_iq_i\\log p_i","definitionContext":"教师不一定是更大的模型，也不一定提供真值"}],"previousBlock":"b-2b528ea2077457439efbcb3f","nextBlock":"b-c5e3b03dbe391904362b875d","prerequisites":[{"id":"b-2b528ea2077457439efbcb3f","url":"#b-2b528ea2077457439efbcb3f","role":"section-title"},{"id":"b-dd468cc7f391ed72b8e5cf90","url":"#b-dd468cc7f391ed72b8e5cf90","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-c5e3b03dbe391904362b875d","type":"p","scope":"block-1861e17c9d7c49f0970e890873be4ff6","url":"https://codingai-lec03.pages.dev/course/notebooks/c0645a4db6e04188ab5b119e87c53bed.html#b-c5e3b03dbe391904362b875d","content":"蒸馏常把软目标项与真实标签项结合，并对温度目标作 $T^2$ 缩放以补偿典型梯度尺度变化。温度不是提高事实正确率的按钮；教师若依赖背景，软分布同样可能传递背景捷径。这里传递的是教师行为中的结构，是否有益需要在目标分布上测量。见 [Hinton 等，2015：蒸馏与温度软目标](https://arxiv.org/abs/1503.02531)。","sha256":"75a4add903cc4a1e6fb297730fba5165423f8e20dbba8243cadb62f713b8be20","chapter":{"id":"I","title":"学习信号与表示"},"unit":{"id":"i-view-objective","title":"视图关系怎样成为学习目标，又会牺牲什么？"},"page":{"id":"reading-block-1861e17c9d7c49f0970e890873be4ff6","title":"教师不一定是更大的模型，也不一定提供真值","number":2},"sectionTitle":"教师不一定是更大的模型，也不一定提供真值","currentQuestion":"教师不一定是更大的模型，也不一定提供真值","sourceSlides":[],"knowledgeBlock":null,"formulas":[{"latex":"T^2","definitionContext":"教师不一定是更大的模型，也不一定提供真值"}],"previousBlock":"b-fb413d3589c141432f2b49c1","nextBlock":"b-3a00eb35cf5acf2e4c59cd44","prerequisites":[{"id":"b-2b528ea2077457439efbcb3f","url":"#b-2b528ea2077457439efbcb3f","role":"section-title"},{"id":"b-dd468cc7f391ed72b8e5cf90","url":"#b-dd468cc7f391ed72b8e5cf90","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-3a00eb35cf5acf2e4c59cd44","type":"p","scope":"block-1861e17c9d7c49f0970e890873be4ff6","url":"https://codingai-lec03.pages.dev/course/notebooks/c0645a4db6e04188ab5b119e87c53bed.html#b-3a00eb35cf5acf2e4c59cd44","content":"自监督教师则可以来自学生自身的历史参数。以 EMA 教师为例，教师参数 $\\xi\\leftarrow m\\xi+(1-m)\\theta$，目标分支停止梯度，学生去预测另一视图的教师输出。这节省人工标签，却没有凭空增加关于类别的权威来源。固定预训练教师、监督蒸馏教师和 EMA 自监督教师的来源与更新方式不同；看到“teacher”一词，先问它由什么数据训练、是否更新、哪条分支接收梯度。","sha256":"792d216843f4f01968151809602fa8c84963defa7cf462797140948c3f044a6b","chapter":{"id":"I","title":"学习信号与表示"},"unit":{"id":"i-view-objective","title":"视图关系怎样成为学习目标，又会牺牲什么？"},"page":{"id":"reading-block-1861e17c9d7c49f0970e890873be4ff6","title":"教师不一定是更大的模型，也不一定提供真值","number":2},"sectionTitle":"教师不一定是更大的模型，也不一定提供真值","currentQuestion":"教师不一定是更大的模型，也不一定提供真值","sourceSlides":[],"knowledgeBlock":null,"formulas":[{"latex":"\\xi\\leftarrow m\\xi+(1-m)\\theta","definitionContext":"教师不一定是更大的模型，也不一定提供真值"}],"previousBlock":"b-c5e3b03dbe391904362b875d","nextBlock":"b-23aeb83178fa4aa532191170","prerequisites":[{"id":"b-2b528ea2077457439efbcb3f","url":"#b-2b528ea2077457439efbcb3f","role":"section-title"},{"id":"b-dd468cc7f391ed72b8e5cf90","url":"#b-dd468cc7f391ed72b8e5cf90","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-23aeb83178fa4aa532191170","type":"table","scope":"block-1861e17c9d7c49f0970e890873be4ff6","url":"https://codingai-lec03.pages.dev/course/notebooks/c0645a4db6e04188ab5b119e87c53bed.html#b-23aeb83178fa4aa532191170","content":"| 设计 | 谁产生目标 | 目标是否随学生训练变化 | 一项必须核对的风险 |\n| --- | --- | --- | --- |\n| 固定教师蒸馏 | 已训练模型 | 通常不变 | 教师错误和偏差会被模仿 |\n| EMA 自监督教师 | 学生参数的移动平均 | 平滑变化 | 常量解或退化解是否受到约束 |\n| 人工类别监督 | 标注规则与标注者 | 数据集内通常固定 | 类别集合遗漏与标注噪声 |","sha256":"5bea10eb69b97fe9ff1f3306fc82766b9958278943003d0ce821c7862c98df24","chapter":{"id":"I","title":"学习信号与表示"},"unit":{"id":"i-view-objective","title":"视图关系怎样成为学习目标，又会牺牲什么？"},"page":{"id":"reading-block-1861e17c9d7c49f0970e890873be4ff6","title":"教师不一定是更大的模型，也不一定提供真值","number":2},"sectionTitle":"教师不一定是更大的模型，也不一定提供真值","currentQuestion":"教师不一定是更大的模型，也不一定提供真值","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"table":{"headers":["设计","谁产生目标","目标是否随学生训练变化","一项必须核对的风险"],"rows":[["设计","谁产生目标","目标是否随学生训练变化","一项必须核对的风险"],["固定教师蒸馏","已训练模型","通常不变","教师错误和偏差会被模仿"],["EMA 自监督教师","学生参数的移动平均","平滑变化","常量解或退化解是否受到约束"],["人工类别监督","标注规则与标注者","数据集内通常固定","类别集合遗漏与标注噪声"]]},"previousBlock":"b-3a00eb35cf5acf2e4c59cd44","nextBlock":"b-eb294f85f5d52b98de0ce3f6","prerequisites":[{"id":"b-2b528ea2077457439efbcb3f","url":"#b-2b528ea2077457439efbcb3f","role":"section-title"},{"id":"b-dd468cc7f391ed72b8e5cf90","url":"#b-dd468cc7f391ed72b8e5cf90","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-eb294f85f5d52b98de0ce3f6","type":"p","scope":"block-1861e17c9d7c49f0970e890873be4ff6","url":"https://codingai-lec03.pages.dev/course/notebooks/c0645a4db6e04188ab5b119e87c53bed.html#b-eb294f85f5d52b98de0ce3f6","content":"BYOL 使用在线/目标网络、预测器和停止梯度等不对称设计；DINO 使用教师—学生分布匹配，并结合 centering、sharpening 等机制。它们给出了不显式使用负例也能学表示的实证路线。[BYOL](https://arxiv.org/abs/2006.07733) · [DINO](https://arxiv.org/abs/2104.14294)。不能把结果压缩成“EMA 单独保证不坍塌”：若随意删掉预测器、改变归一化或温度，系统已不是论文中的训练组合；只观察训练损失也识别不了所有退化。","sha256":"ad1c9cdc8c931315cefa1f74029155573a31c600a5d190e736b1616f4743d31a","chapter":{"id":"I","title":"学习信号与表示"},"unit":{"id":"i-view-objective","title":"视图关系怎样成为学习目标，又会牺牲什么？"},"page":{"id":"reading-block-1861e17c9d7c49f0970e890873be4ff6","title":"教师不一定是更大的模型，也不一定提供真值","number":2},"sectionTitle":"教师不一定是更大的模型，也不一定提供真值","currentQuestion":"教师不一定是更大的模型，也不一定提供真值","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-23aeb83178fa4aa532191170","nextBlock":null,"prerequisites":[{"id":"b-2b528ea2077457439efbcb3f","url":"#b-2b528ea2077457439efbcb3f","role":"section-title"},{"id":"b-dd468cc7f391ed72b8e5cf90","url":"#b-dd468cc7f391ed72b8e5cf90","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-cd1df5b97055d6e0f9dd91ee","type":"h2","scope":"block-05c08a06e13d4bcb9ad2855d108f9310","url":"https://codingai-lec03.pages.dev/course/notebooks/c0645a4db6e04188ab5b119e87c53bed.html#b-cd1df5b97055d6e0f9dd91ee","content":"## 先用三个量区分坍塌与无用读出","sha256":"7b046569644a82f65ad052205227f16c1b004172e34365cbe4de73c52681b420","chapter":{"id":"I","title":"学习信号与表示"},"unit":{"id":"i-view-objective","title":"视图关系怎样成为学习目标，又会牺牲什么？"},"page":{"id":"reading-block-05c08a06e13d4bcb9ad2855d108f9310","title":"先用三个量区分坍塌与无用读出","number":3},"sectionTitle":"先用三个量区分坍塌与无用读出","currentQuestion":"先用三个量区分坍塌与无用读出","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":null,"nextBlock":"b-1051bf024f760b86e31743c8","prerequisites":[{"id":"b-1051bf024f760b86e31743c8","url":"#b-1051bf024f760b86e31743c8","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-1051bf024f760b86e31743c8","type":"p","scope":"block-05c08a06e13d4bcb9ad2855d108f9310","url":"https://codingai-lec03.pages.dev/course/notebooks/c0645a4db6e04188ab5b119e87c53bed.html#b-1051bf024f760b86e31743c8","content":"若一批表示全部为 $(1,1)$，协方差是零，任意两个样本距离为零；仅凭这些表示的分类器只能按先验猜类别。若表示分布有方差，仍可能仅保存背景亮度，不能因“看起来散开了”就判断语义有用。相反，线性分类失败也不必说明标签信息消失。","sha256":"3c4a62216e8f3c52833f83e5a8ceb945b5bdb027d0b6deabca56ad1fba8c71e9","chapter":{"id":"I","title":"学习信号与表示"},"unit":{"id":"i-view-objective","title":"视图关系怎样成为学习目标，又会牺牲什么？"},"page":{"id":"reading-block-05c08a06e13d4bcb9ad2855d108f9310","title":"先用三个量区分坍塌与无用读出","number":3},"sectionTitle":"先用三个量区分坍塌与无用读出","currentQuestion":"先用三个量区分坍塌与无用读出","sourceSlides":[],"knowledgeBlock":null,"formulas":[{"latex":"(1,1)","definitionContext":"先用三个量区分坍塌与无用读出"}],"previousBlock":"b-cd1df5b97055d6e0f9dd91ee","nextBlock":"b-02820f378c7ad319163233f4","prerequisites":[{"id":"b-cd1df5b97055d6e0f9dd91ee","url":"#b-cd1df5b97055d6e0f9dd91ee","role":"section-title"}],"relatedBlocks":[]},{"id":"b-02820f378c7ad319163233f4","type":"p","scope":"block-05c08a06e13d4bcb9ad2855d108f9310","url":"https://codingai-lec03.pages.dev/course/notebooks/c0645a4db6e04188ab5b119e87c53bed.html#b-02820f378c7ad319163233f4","content":"用四个等概率输入 $h\\in\\{(-1,-1),(-1,1),(1,-1),(1,1)\\}$，同号标签 0，异号标签 1。这是 XOR：四点没有碰撞，每一点都唯一确定标签，但两类凸包都经过原点，不能被一条直线严格分开。加入 $z=h_1h_2$ 后，标签为 $\\mathbf1[z<0]$，一维阈值就能完全分开。非线性变换没有创造标签信息，只改变了可读出形式。","sha256":"5cfb5233b27d8709e56ed0dc88b6b62514c0f383716469d139b170d4beb2f439","chapter":{"id":"I","title":"学习信号与表示"},"unit":{"id":"i-view-objective","title":"视图关系怎样成为学习目标，又会牺牲什么？"},"page":{"id":"reading-block-05c08a06e13d4bcb9ad2855d108f9310","title":"先用三个量区分坍塌与无用读出","number":3},"sectionTitle":"先用三个量区分坍塌与无用读出","currentQuestion":"先用三个量区分坍塌与无用读出","sourceSlides":[],"knowledgeBlock":null,"formulas":[{"latex":"h\\in\\{(-1,-1),(-1,1),(1,-1),(1,1)\\}","definitionContext":"先用三个量区分坍塌与无用读出"},{"latex":"z=h_1h_2","definitionContext":"先用三个量区分坍塌与无用读出"},{"latex":"\\mathbf1[z<0]","definitionContext":"先用三个量区分坍塌与无用读出"}],"previousBlock":"b-1051bf024f760b86e31743c8","nextBlock":"b-74dc730edd62ab4322618a76","prerequisites":[{"id":"b-cd1df5b97055d6e0f9dd91ee","url":"#b-cd1df5b97055d6e0f9dd91ee","role":"section-title"},{"id":"b-1051bf024f760b86e31743c8","url":"#b-1051bf024f760b86e31743c8","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-74dc730edd62ab4322618a76","type":"p","scope":"block-05c08a06e13d4bcb9ad2855d108f9310","url":"https://codingai-lec03.pages.dev/course/notebooks/c0645a4db6e04188ab5b119e87c53bed.html#b-74dc730edd62ab4322618a76","content":"![XOR 保留信息但线性不可分，以及域变化中的两种线索](/course/repair/assets/cls07-xor-domain.svg)","sha256":"3dea9621641a48772801e1ca7f637ce261694961d4996c34799d658eb0295c99","chapter":{"id":"I","title":"学习信号与表示"},"unit":{"id":"i-view-objective","title":"视图关系怎样成为学习目标，又会牺牲什么？"},"page":{"id":"reading-block-05c08a06e13d4bcb9ad2855d108f9310","title":"先用三个量区分坍塌与无用读出","number":3},"sectionTitle":"先用三个量区分坍塌与无用读出","currentQuestion":"先用三个量区分坍塌与无用读出","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-02820f378c7ad319163233f4","nextBlock":"b-2586af661eb44d4b4ae610ad","prerequisites":[{"id":"b-cd1df5b97055d6e0f9dd91ee","url":"#b-cd1df5b97055d6e0f9dd91ee","role":"section-title"},{"id":"b-1051bf024f760b86e31743c8","url":"#b-1051bf024f760b86e31743c8","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-2586af661eb44d4b4ae610ad","type":"p","scope":"block-05c08a06e13d4bcb9ad2855d108f9310","url":"https://codingai-lec03.pages.dev/course/notebooks/c0645a4db6e04188ab5b119e87c53bed.html#b-2586af661eb44d4b4ae610ad","content":"左图四点的身份与标签均完整保留，却没有线性分界；右图给出一个人为设定的环境分布：颜色线索从 90% 可靠变为 10%，形状线索保持 80%。这些是解释条件的构造数值，不是实测模型成绩。","sha256":"b8b70ff82c0bcd32ddaf4e44dea0f80f7c67697b267456c27d17cf9879c89401","chapter":{"id":"I","title":"学习信号与表示"},"unit":{"id":"i-view-objective","title":"视图关系怎样成为学习目标，又会牺牲什么？"},"page":{"id":"reading-block-05c08a06e13d4bcb9ad2855d108f9310","title":"先用三个量区分坍塌与无用读出","number":3},"sectionTitle":"先用三个量区分坍塌与无用读出","currentQuestion":"先用三个量区分坍塌与无用读出","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-74dc730edd62ab4322618a76","nextBlock":"b-03d863e5189e6836edf94b2a","prerequisites":[{"id":"b-cd1df5b97055d6e0f9dd91ee","url":"#b-cd1df5b97055d6e0f9dd91ee","role":"section-title"},{"id":"b-1051bf024f760b86e31743c8","url":"#b-1051bf024f760b86e31743c8","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-03d863e5189e6836edf94b2a","type":"p","scope":"block-05c08a06e13d4bcb9ad2855d108f9310","url":"https://codingai-lec03.pages.dev/course/notebooks/c0645a4db6e04188ab5b119e87c53bed.html#b-03d863e5189e6836edf94b2a","content":"因此值得分别检查表示方差/秩、冻结读出、允许非线性适配的成绩。每一种检查排除的解释不同。[LAB 17](https://deepnote.com/project/38755e4e-23f6-4f5f-8c68-8a2e4114b1c1/notebook/268a4ab11b464c64b065f9fc12b19cc7) 对照软目标和常量表示；[LAB 18](https://deepnote.com/project/38755e4e-23f6-4f5f-8c68-8a2e4114b1c1/notebook/33066318f5884adc8417068df0f40431) 则比较信息是否保留与读出是否容易。不要把前者的一次低距离当成后者的高质量表示。","sha256":"30c4d3d14ad50393c706ffbf805526ef1d5433c63b4d438f68c0a798cc524ae6","chapter":{"id":"I","title":"学习信号与表示"},"unit":{"id":"i-view-objective","title":"视图关系怎样成为学习目标，又会牺牲什么？"},"page":{"id":"reading-block-05c08a06e13d4bcb9ad2855d108f9310","title":"先用三个量区分坍塌与无用读出","number":3},"sectionTitle":"先用三个量区分坍塌与无用读出","currentQuestion":"先用三个量区分坍塌与无用读出","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-2586af661eb44d4b4ae610ad","nextBlock":"b-d16040819ba81a3bdfe0bf80","prerequisites":[{"id":"b-cd1df5b97055d6e0f9dd91ee","url":"#b-cd1df5b97055d6e0f9dd91ee","role":"section-title"},{"id":"b-1051bf024f760b86e31743c8","url":"#b-1051bf024f760b86e31743c8","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-d16040819ba81a3bdfe0bf80","type":"h2","scope":"block-05c08a06e13d4bcb9ad2855d108f9310","url":"https://codingai-lec03.pages.dev/course/notebooks/c0645a4db6e04188ab5b119e87c53bed.html#b-d16040819ba81a3bdfe0bf80","content":"## 数据处理不等式究竟限制哪一条链","sha256":"f297f61db6a828793849fb1c0a83af36c1b9e862d1024cebda5de5f7fcdee6ff","chapter":{"id":"I","title":"学习信号与表示"},"unit":{"id":"i-view-objective","title":"视图关系怎样成为学习目标，又会牺牲什么？"},"page":{"id":"reading-block-05c08a06e13d4bcb9ad2855d108f9310","title":"先用三个量区分坍塌与无用读出","number":3},"sectionTitle":"先用三个量区分坍塌与无用读出","currentQuestion":"先用三个量区分坍塌与无用读出","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-03d863e5189e6836edf94b2a","nextBlock":"b-d0db3b6e0edd6a6be0672c22","prerequisites":[{"id":"b-cd1df5b97055d6e0f9dd91ee","url":"#b-cd1df5b97055d6e0f9dd91ee","role":"section-title"},{"id":"b-1051bf024f760b86e31743c8","url":"#b-1051bf024f760b86e31743c8","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-d0db3b6e0edd6a6be0672c22","type":"p","scope":"block-05c08a06e13d4bcb9ad2855d108f9310","url":"https://codingai-lec03.pages.dev/course/notebooks/c0645a4db6e04188ab5b119e87c53bed.html#b-d0db3b6e0edd6a6be0672c22","content":"设 $X=(A,B)$ 为两个独立均匀 bit，标签 $Y=A$。固定表示 $H_1=A$ 只存一个 bit，却保留全部标签信息：$I(Y;H_1)=1$ bit。另一个一 bit 表示 $H_2=A\\oplus B$ 与 $Y$ 独立，所以 $I(Y;H_2)=0$。两者维数相同、熵相同，任务信息却不同。这比“压缩越强越好”更能说明问题。","sha256":"0fcfc28644b12ce434e17f07aa61a599852cad27f630da57f20efd4c7ad4037d","chapter":{"id":"I","title":"学习信号与表示"},"unit":{"id":"i-view-objective","title":"视图关系怎样成为学习目标，又会牺牲什么？"},"page":{"id":"reading-block-05c08a06e13d4bcb9ad2855d108f9310","title":"先用三个量区分坍塌与无用读出","number":3},"sectionTitle":"先用三个量区分坍塌与无用读出","currentQuestion":"先用三个量区分坍塌与无用读出","sourceSlides":[],"knowledgeBlock":null,"formulas":[{"latex":"X=(A,B)","definitionContext":"先用三个量区分坍塌与无用读出"},{"latex":"Y=A","definitionContext":"先用三个量区分坍塌与无用读出"},{"latex":"H_1=A","definitionContext":"先用三个量区分坍塌与无用读出"},{"latex":"I(Y;H_1)=1","definitionContext":"先用三个量区分坍塌与无用读出"},{"latex":"H_2=A\\oplus B","definitionContext":"先用三个量区分坍塌与无用读出"},{"latex":"Y","definitionContext":"先用三个量区分坍塌与无用读出"},{"latex":"I(Y;H_2)=0","definitionContext":"先用三个量区分坍塌与无用读出"}],"previousBlock":"b-d16040819ba81a3bdfe0bf80","nextBlock":"b-26330905fc670d466b3f9e7c","prerequisites":[{"id":"b-d16040819ba81a3bdfe0bf80","url":"#b-d16040819ba81a3bdfe0bf80","role":"section-title"},{"id":"b-1051bf024f760b86e31743c8","url":"#b-1051bf024f760b86e31743c8","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-26330905fc670d466b3f9e7c","type":"p","scope":"block-05c08a06e13d4bcb9ad2855d108f9310","url":"https://codingai-lec03.pages.dev/course/notebooks/c0645a4db6e04188ab5b119e87c53bed.html#b-26330905fc670d466b3f9e7c","content":"在固定的 Markov 链 $Y\\to X\\to H$ 中，数据处理不等式给出 $I(Y;H)\\le I(Y;X)$：后处理不能从已经完全相同的输入状态中恢复被删除的区别。其前提是表示由给定 $X$ 产生，没有额外关于 $Y$ 的信息通道。模型训练时参数 $\\theta$ 可从外部训练集获得规律；比较训练前后的 $f_\\theta$，并不是在同一条固定前向链上凭空创造信息。连续变量的互信息还可能因确定性映射出现技术问题，不能拿离散 bit 算例的数值直接衡量一张连续特征图。课程采用这个例子是说明条件，而非用互信息替代实际迁移评价。","sha256":"aa3bbe0b7cc6c328d10bb19ad23b01d3e3c11498cbcd139289bbbc11c3e6c9a1","chapter":{"id":"I","title":"学习信号与表示"},"unit":{"id":"i-view-objective","title":"视图关系怎样成为学习目标，又会牺牲什么？"},"page":{"id":"reading-block-05c08a06e13d4bcb9ad2855d108f9310","title":"先用三个量区分坍塌与无用读出","number":3},"sectionTitle":"先用三个量区分坍塌与无用读出","currentQuestion":"先用三个量区分坍塌与无用读出","sourceSlides":[],"knowledgeBlock":null,"formulas":[{"latex":"Y\\to X\\to H","definitionContext":"先用三个量区分坍塌与无用读出"},{"latex":"I(Y;H)\\le I(Y;X)","definitionContext":"先用三个量区分坍塌与无用读出"},{"latex":"X","definitionContext":"先用三个量区分坍塌与无用读出"},{"latex":"Y","definitionContext":"先用三个量区分坍塌与无用读出"},{"latex":"\\theta","definitionContext":"先用三个量区分坍塌与无用读出"},{"latex":"f_\\theta","definitionContext":"先用三个量区分坍塌与无用读出"}],"previousBlock":"b-d0db3b6e0edd6a6be0672c22","nextBlock":null,"prerequisites":[{"id":"b-d16040819ba81a3bdfe0bf80","url":"#b-d16040819ba81a3bdfe0bf80","role":"section-title"},{"id":"b-1051bf024f760b86e31743c8","url":"#b-1051bf024f760b86e31743c8","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-c6ee3368384109f6a605e6ef","type":"h2","scope":"block-0d856b8e3bd442f9bd30fd3a2b3a9d60","url":"https://codingai-lec03.pages.dev/course/notebooks/c0645a4db6e04188ab5b119e87c53bed.html#b-c6ee3368384109f6a605e6ef","content":"## 分布变了，源域最方便的线索可能最危险","sha256":"718136f0278e8a0ff5a30f1c38d5890c1f388a5bd212c24222a2e8e229889387","chapter":{"id":"I","title":"学习信号与表示"},"unit":{"id":"i-view-objective","title":"视图关系怎样成为学习目标，又会牺牲什么？"},"page":{"id":"reading-block-0d856b8e3bd442f9bd30fd3a2b3a9d60","title":"分布变了，源域最方便的线索可能最危险","number":4},"sectionTitle":"分布变了，源域最方便的线索可能最危险","currentQuestion":"分布变了，源域最方便的线索可能最危险","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":null,"nextBlock":"b-28c611885c13f84f89e1ae3e","prerequisites":[{"id":"b-28c611885c13f84f89e1ae3e","url":"#b-28c611885c13f84f89e1ae3e","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-28c611885c13f84f89e1ae3e","type":"p","scope":"block-0d856b8e3bd442f9bd30fd3a2b3a9d60","url":"https://codingai-lec03.pages.dev/course/notebooks/c0645a4db6e04188ab5b119e87c53bed.html#b-28c611885c13f84f89e1ae3e","content":"考虑二分类图像：训练环境中，颜色与标签一致率为 90%，形状为 80%；新环境中颜色一致率降到 10%，形状仍为 80%。一个只用颜色的规则在源域期望准确率为 90%、目标域为 10%；只用形状则两域都为 80%。这里没有真正训练网络，因此只能证明规则在指定分布下失效，不能声称已经测得迁移带来的负收益。","sha256":"60c3e09851226bcb5159f1fda0394bde192080355322a0673534905c9090a271","chapter":{"id":"I","title":"学习信号与表示"},"unit":{"id":"i-view-objective","title":"视图关系怎样成为学习目标，又会牺牲什么？"},"page":{"id":"reading-block-0d856b8e3bd442f9bd30fd3a2b3a9d60","title":"分布变了，源域最方便的线索可能最危险","number":4},"sectionTitle":"分布变了，源域最方便的线索可能最危险","currentQuestion":"分布变了，源域最方便的线索可能最危险","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-c6ee3368384109f6a605e6ef","nextBlock":"b-a1d3a29c5b5b7f81185c49ff","prerequisites":[{"id":"b-c6ee3368384109f6a605e6ef","url":"#b-c6ee3368384109f6a605e6ef","role":"section-title"}],"relatedBlocks":[]},{"id":"b-a1d3a29c5b5b7f81185c49ff","type":"p","scope":"block-0d856b8e3bd442f9bd30fd3a2b3a9d60","url":"https://codingai-lec03.pages.dev/course/notebooks/c0645a4db6e04188ab5b119e87c53bed.html#b-a1d3a29c5b5b7f81185c49ff","content":"要声称**负迁移**，需要同一目标任务、标签预算、训练和评价划分下，比较源预训练初始化与合理的目标域基线，例如从头训练或其他初始化；若前者更差，差值才是当前协议下的迁移负效应。冻结、局部微调、全量微调允许不同程度改写源线索，不能仅比较训练准确率。若目标标签很少，全量更新还可能把小样本噪声写进整个网络。","sha256":"b655be870fde80e77afd34b140e3a1a1938336d19183cb649e78ee4e026567d1","chapter":{"id":"I","title":"学习信号与表示"},"unit":{"id":"i-view-objective","title":"视图关系怎样成为学习目标，又会牺牲什么？"},"page":{"id":"reading-block-0d856b8e3bd442f9bd30fd3a2b3a9d60","title":"分布变了，源域最方便的线索可能最危险","number":4},"sectionTitle":"分布变了，源域最方便的线索可能最危险","currentQuestion":"分布变了，源域最方便的线索可能最危险","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-28c611885c13f84f89e1ae3e","nextBlock":"b-8cc7eba8293e6e1d91eaf1e0","prerequisites":[{"id":"b-c6ee3368384109f6a605e6ef","url":"#b-c6ee3368384109f6a605e6ef","role":"section-title"},{"id":"b-28c611885c13f84f89e1ae3e","url":"#b-28c611885c13f84f89e1ae3e","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-8cc7eba8293e6e1d91eaf1e0","type":"p","scope":"block-0d856b8e3bd442f9bd30fd3a2b3a9d60","url":"https://codingai-lec03.pages.dev/course/notebooks/c0645a4db6e04188ab5b119e87c53bed.html#b-8cc7eba8293e6e1d91eaf1e0","content":"随机划分若同时把同一背景、拍摄设备或主体身份分进两侧，可能隐藏域变化。按环境或主体分组的保留集更接近“到新场景使用”的问题；验证集仍须用于选择适配强度，不能看完测试环境再挑出最好的一种方法。[D2L：环境与分布偏移](https://d2l.ai/chapter_linear-classification/environment-and-distribution-shift.html) 给出这些假设变化的基础讨论。[LAB 19](https://deepnote.com/workspace/ise3309jingao-5ba66b0a-4283-4f5a-ae06-3f68d3201e99/project/38755e4e-23f6-4f5f-8c68-8a2e4114b1c1/notebook/c20ef6a65d184331b1b52bf351077504?utm_source=openai&utm_medium=mcp&utm_campaign=openaimcp&utm_content=c20ef6a65d184331b1b52bf351077504&utm_term=get_notebook) 的真实逻辑回归与适配比较（输入为两维解析特征，不是照片）和 [LAB 02](https://deepnote.com/workspace/ise3309jingao-5ba66b0a-4283-4f5a-ae06-3f68d3201e99/project/38755e4e-23f6-4f5f-8c68-8a2e4114b1c1/notebook/4bd450d54870432db91c80a4df563b8e?utm_source=openai&utm_medium=mcp&utm_campaign=openaimcp&utm_content=4bd450d54870432db91c80a4df563b8e&utm_term=get_notebook) 的实际参数迁移各承担其中一层证据。","sha256":"d622785e68c1be46b0cab51204b3da00c5c09a4111af63a00a58898ce3ad502c","chapter":{"id":"I","title":"学习信号与表示"},"unit":{"id":"i-view-objective","title":"视图关系怎样成为学习目标，又会牺牲什么？"},"page":{"id":"reading-block-0d856b8e3bd442f9bd30fd3a2b3a9d60","title":"分布变了，源域最方便的线索可能最危险","number":4},"sectionTitle":"分布变了，源域最方便的线索可能最危险","currentQuestion":"分布变了，源域最方便的线索可能最危险","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-a1d3a29c5b5b7f81185c49ff","nextBlock":"b-6d9778373e0abd007e832054","prerequisites":[{"id":"b-c6ee3368384109f6a605e6ef","url":"#b-c6ee3368384109f6a605e6ef","role":"section-title"},{"id":"b-28c611885c13f84f89e1ae3e","url":"#b-28c611885c13f84f89e1ae3e","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-6d9778373e0abd007e832054","type":"p","scope":"block-0d856b8e3bd442f9bd30fd3a2b3a9d60","url":"https://codingai-lec03.pages.dev/course/notebooks/c0645a4db6e04188ab5b119e87c53bed.html#b-6d9778373e0abd007e832054","content":"第一条主线到这里形成一个闭环：重建规定要保留什么，对比规定哪些变化要忽略，教师规定向谁靠近，而读出与新域评价检查这些选择是否支持使用。下一册 [CLS 07](https://deepnote.com/workspace/ise3309jingao-5ba66b0a-4283-4f5a-ae06-3f68d3201e99/project/38755e4e-23f6-4f5f-8c68-8a2e4114b1c1/notebook/586e66e049764158a50747e69c219312?utm_source=openai&utm_medium=mcp&utm_campaign=openaimcp&utm_content=586e66e049764158a50747e69c219312&utm_term=get_notebook) 改变的不是标签供给，而是输出要求：即使整图类别判断正确，网络还要交付对象在哪里。","sha256":"12bcd5d53dcdcdd3a25813b2723f5fe32ed93be252405fcf1d9df668daa7d8a8","chapter":{"id":"I","title":"学习信号与表示"},"unit":{"id":"i-view-objective","title":"视图关系怎样成为学习目标，又会牺牲什么？"},"page":{"id":"reading-block-0d856b8e3bd442f9bd30fd3a2b3a9d60","title":"分布变了，源域最方便的线索可能最危险","number":4},"sectionTitle":"分布变了，源域最方便的线索可能最危险","currentQuestion":"分布变了，源域最方便的线索可能最危险","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-8cc7eba8293e6e1d91eaf1e0","nextBlock":null,"prerequisites":[{"id":"b-c6ee3368384109f6a605e6ef","url":"#b-c6ee3368384109f6a605e6ef","role":"section-title"},{"id":"b-28c611885c13f84f89e1ae3e","url":"#b-28c611885c13f84f89e1ae3e","role":"problem-context"}],"relatedBlocks":[]}]}