{"version":2,"relationshipURLBase":"https://codingai-lec03.pages.dev/course/blocks/source-task-loss-three-axes.html","relationshipResolution":"Resolve relative relationship URLs against relationshipURLBase; complete records and absolute URLs are in blocks.","title":"监督来源、预测任务与损失要分开","url":"https://codingai-lec03.pages.dev/course/blocks/source-task-loss-three-axes.html","markdown":"https://codingai-lec03.pages.dev/course/blocks/source-task-loss-three-axes.md","pages":[{"id":"reading-opening","title":"监督来源、预测任务与损失要分开","unit":"训练留下什么，训练答案又从哪里来？","unitId":"i-reuse-and-targets","chapter":"I","number":1}],"aliases":[{"oldId":"b-78c3ef3aa77e1e1f89478e9d","targetId":"b-78c3ef3aa77e1e1f89478e9d","page":"reading-opening","operation":"move","beforeSHA256":"c5f31347709dd5e06d2569b7ec002b5609e847f62e6aeafd0d5284d8f164daf2","afterSHA256":"c5f31347709dd5e06d2569b7ec002b5609e847f62e6aeafd0d5284d8f164daf2","reason":"Same source body in authored semantic page"},{"oldId":"b-3138a07fb78ae80e74ca2972","targetId":"b-3138a07fb78ae80e74ca2972","page":"reading-opening","operation":"move","beforeSHA256":"44fcc603fecb14d20e23aa048ec8a6c1eb60be204c2985f3c11a8fcad8b3e0cc","afterSHA256":"44fcc603fecb14d20e23aa048ec8a6c1eb60be204c2985f3c11a8fcad8b3e0cc","reason":"Same source body in authored semantic page"},{"oldId":"b-05a4cf1d0da83e72b0a2f702","targetId":"b-05a4cf1d0da83e72b0a2f702","page":"reading-opening","operation":"move","beforeSHA256":"7ac6d742d7f20aefddd91a3079f15540f63407a1a49423b0c9a537207a0c4df0","afterSHA256":"7ac6d742d7f20aefddd91a3079f15540f63407a1a49423b0c9a537207a0c4df0","reason":"Same source body in authored semantic page"},{"oldId":"b-0e8b31e40489a85b3dd04e86","targetId":"b-0e8b31e40489a85b3dd04e86","page":"reading-opening","operation":"move","beforeSHA256":"f0af355e8a3f5aa0c0f7113e97480a6ba1bace972a6506476e267b34a1f40fad","afterSHA256":"f0af355e8a3f5aa0c0f7113e97480a6ba1bace972a6506476e267b34a1f40fad","reason":"Same source body in authored semantic page"},{"oldId":"b-b13a38972a504a75cfc807ca","targetId":"b-b13a38972a504a75cfc807ca","page":"reading-opening","operation":"move","beforeSHA256":"aa336435506b60510a3d1586b97f28e0b141ae4fb9d131923ce3b9a09d026630","afterSHA256":"aa336435506b60510a3d1586b97f28e0b141ae4fb9d131923ce3b9a09d026630","reason":"Same source body in authored semantic page"},{"oldId":"b-46dd59219cc3676d60f36e6b","targetId":"b-46dd59219cc3676d60f36e6b","page":"reading-opening","operation":"move","beforeSHA256":"2b9ccef7e1220e2250cceff1abd8bdd736b42f8da5a223f26b029f44b962e807","afterSHA256":"2b9ccef7e1220e2250cceff1abd8bdd736b42f8da5a223f26b029f44b962e807","reason":"Same source body in authored semantic page"}],"blocks":[{"id":"b-78c3ef3aa77e1e1f89478e9d","type":"h1","scope":"source-task-loss-three-axes","url":"https://codingai-lec03.pages.dev/course/blocks/source-task-loss-three-axes.html#b-78c3ef3aa77e1e1f89478e9d","content":"# 监督来源、预测任务与损失要分开","sha256":"c5f31347709dd5e06d2569b7ec002b5609e847f62e6aeafd0d5284d8f164daf2","chapter":{"id":"I","title":"学习信号与表示"},"unit":{"id":"i-reuse-and-targets","title":"训练留下什么，训练答案又从哪里来？"},"page":{"id":"reading-opening","title":"监督来源、预测任务与损失要分开","number":1},"sectionTitle":"监督来源、预测任务与损失要分开","currentQuestion":"监督来源、预测任务与损失要分开","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":null,"nextBlock":"b-3138a07fb78ae80e74ca2972","prerequisites":[{"id":"b-3138a07fb78ae80e74ca2972","url":"#b-3138a07fb78ae80e74ca2972","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-3138a07fb78ae80e74ca2972","type":"p","scope":"source-task-loss-three-axes","url":"https://codingai-lec03.pages.dev/course/blocks/source-task-loss-three-axes.html#b-3138a07fb78ae80e74ca2972","content":"把“自监督、分类、交叉熵”放回不同维度，后面才不会用方法名替代完整学习链。","sha256":"44fcc603fecb14d20e23aa048ec8a6c1eb60be204c2985f3c11a8fcad8b3e0cc","chapter":{"id":"I","title":"学习信号与表示"},"unit":{"id":"i-reuse-and-targets","title":"训练留下什么，训练答案又从哪里来？"},"page":{"id":"reading-opening","title":"监督来源、预测任务与损失要分开","number":1},"sectionTitle":"监督来源、预测任务与损失要分开","currentQuestion":"监督来源、预测任务与损失要分开","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-78c3ef3aa77e1e1f89478e9d","nextBlock":"b-05a4cf1d0da83e72b0a2f702","prerequisites":[{"id":"b-78c3ef3aa77e1e1f89478e9d","url":"#b-78c3ef3aa77e1e1f89478e9d","role":"section-title"}],"relatedBlocks":[]},{"id":"b-05a4cf1d0da83e72b0a2f702","type":"p","scope":"source-task-loss-three-axes","url":"https://codingai-lec03.pages.dev/course/blocks/source-task-loss-three-axes.html#b-05a4cf1d0da83e72b0a2f702","content":"“自监督”说明目标怎样取得；“分类”说明输出怎样组织；“交叉熵”说明怎样惩罚预测。这是三个维度，不是互斥方法名。人工类别标签可以训练softmax分类，从同图视图身份构造的标签也可以训练softmax分类；两者都能用交叉熵。","sha256":"7ac6d742d7f20aefddd91a3079f15540f63407a1a49423b0c9a537207a0c4df0","chapter":{"id":"I","title":"学习信号与表示"},"unit":{"id":"i-reuse-and-targets","title":"训练留下什么，训练答案又从哪里来？"},"page":{"id":"reading-opening","title":"监督来源、预测任务与损失要分开","number":1},"sectionTitle":"监督来源、预测任务与损失要分开","currentQuestion":"监督来源、预测任务与损失要分开","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-3138a07fb78ae80e74ca2972","nextBlock":"b-0e8b31e40489a85b3dd04e86","prerequisites":[{"id":"b-78c3ef3aa77e1e1f89478e9d","url":"#b-78c3ef3aa77e1e1f89478e9d","role":"section-title"},{"id":"b-3138a07fb78ae80e74ca2972","url":"#b-3138a07fb78ae80e74ca2972","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-0e8b31e40489a85b3dd04e86","type":"p","scope":"source-task-loss-three-axes","url":"https://codingai-lec03.pages.dev/course/blocks/source-task-loss-three-axes.html#b-0e8b31e40489a85b3dd04e86","content":"对比学习的一题是“候选中哪一个和查询同源”；遮挡重建的一题是“这个隐藏位置原来是什么像素”。它们都从数据构造目标，却要求保留不同结构。目标不是“没有答案”，只是答案不是额外写下的类别名称。","sha256":"f0af355e8a3f5aa0c0f7113e97480a6ba1bace972a6506476e267b34a1f40fad","chapter":{"id":"I","title":"学习信号与表示"},"unit":{"id":"i-reuse-and-targets","title":"训练留下什么，训练答案又从哪里来？"},"page":{"id":"reading-opening","title":"监督来源、预测任务与损失要分开","number":1},"sectionTitle":"监督来源、预测任务与损失要分开","currentQuestion":"监督来源、预测任务与损失要分开","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-05a4cf1d0da83e72b0a2f702","nextBlock":"b-b13a38972a504a75cfc807ca","prerequisites":[{"id":"b-78c3ef3aa77e1e1f89478e9d","url":"#b-78c3ef3aa77e1e1f89478e9d","role":"section-title"},{"id":"b-3138a07fb78ae80e74ca2972","url":"#b-3138a07fb78ae80e74ca2972","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-b13a38972a504a75cfc807ca","type":"p","scope":"source-task-loss-three-axes","url":"https://codingai-lec03.pages.dev/course/blocks/source-task-loss-three-axes.html#b-b13a38972a504a75cfc807ca","content":"判断一个实验时，依次写出输入、目标的取得过程、模型输出、损失和最终评价。若只写“用了自监督”，还不知道它学到了什么；若写出完整链条，就能发现有没有把隐藏真值意外送入输入，或把训练损失当成另一项任务的成绩。","sha256":"aa336435506b60510a3d1586b97f28e0b141ae4fb9d131923ce3b9a09d026630","chapter":{"id":"I","title":"学习信号与表示"},"unit":{"id":"i-reuse-and-targets","title":"训练留下什么，训练答案又从哪里来？"},"page":{"id":"reading-opening","title":"监督来源、预测任务与损失要分开","number":1},"sectionTitle":"监督来源、预测任务与损失要分开","currentQuestion":"监督来源、预测任务与损失要分开","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-0e8b31e40489a85b3dd04e86","nextBlock":"b-46dd59219cc3676d60f36e6b","prerequisites":[{"id":"b-78c3ef3aa77e1e1f89478e9d","url":"#b-78c3ef3aa77e1e1f89478e9d","role":"section-title"},{"id":"b-3138a07fb78ae80e74ca2972","url":"#b-3138a07fb78ae80e74ca2972","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-46dd59219cc3676d60f36e6b","type":"h2","scope":"source-task-loss-three-axes","url":"https://codingai-lec03.pages.dev/course/blocks/source-task-loss-three-axes.html#b-46dd59219cc3676d60f36e6b","content":"## 继续阅读与实验","sha256":"2b9ccef7e1220e2250cceff1abd8bdd736b42f8da5a223f26b029f44b962e807","chapter":{"id":"I","title":"学习信号与表示"},"unit":{"id":"i-reuse-and-targets","title":"训练留下什么，训练答案又从哪里来？"},"page":{"id":"reading-opening","title":"监督来源、预测任务与损失要分开","number":1},"sectionTitle":"监督来源、预测任务与损失要分开","currentQuestion":"监督来源、预测任务与损失要分开","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-b13a38972a504a75cfc807ca","nextBlock":null,"prerequisites":[{"id":"b-78c3ef3aa77e1e1f89478e9d","url":"#b-78c3ef3aa77e1e1f89478e9d","role":"section-title"},{"id":"b-3138a07fb78ae80e74ca2972","url":"#b-3138a07fb78ae80e74ca2972","role":"problem-context"}],"relatedBlocks":[]}]}