{"version":2,"relationshipURLBase":"https://codingai-lec03.pages.dev/course/blocks/caption-decoding-later.html","relationshipResolution":"Resolve relative relationship URLs against relationshipURLBase; complete records and absolute URLs are in blocks.","title":"选读：局部最高概率为何不保证整句最高概率","url":"https://codingai-lec03.pages.dev/course/blocks/caption-decoding-later.html","markdown":"https://codingai-lec03.pages.dev/course/blocks/caption-decoding-later.md","pages":[{"id":"reading-opening","title":"选读：局部最高概率为何不保证整句最高概率","unit":"一个条件怎样容纳多种合理结果？","unitId":"generation-many-possible-results","chapter":"III","number":1}],"aliases":[{"oldId":"b-5336bea76536cfaddfdb696c","targetId":"b-5336bea76536cfaddfdb696c","page":"reading-opening","operation":"move","beforeSHA256":"07ef2b2a2dc1aab1aacdb60c913d70746a882f51b820c051f18b7d1c30021afd","afterSHA256":"07ef2b2a2dc1aab1aacdb60c913d70746a882f51b820c051f18b7d1c30021afd","reason":"Same source body in authored semantic page"},{"oldId":"b-289caffc80977c5811065725","targetId":"b-289caffc80977c5811065725","page":"reading-opening","operation":"move","beforeSHA256":"513b4dde07c19aa85a702ba5af2c592cb8b134338c12e49e6c6f6faf9adf92ed","afterSHA256":"513b4dde07c19aa85a702ba5af2c592cb8b134338c12e49e6c6f6faf9adf92ed","reason":"Same source body in authored semantic page"},{"oldId":"b-6a09f2304f5fd641979b4394","targetId":"b-6a09f2304f5fd641979b4394","page":"reading-opening","operation":"move","beforeSHA256":"aeaad29331bc2c94ac240ece646024336e7455de08d754b00ed438164663d9be","afterSHA256":"aeaad29331bc2c94ac240ece646024336e7455de08d754b00ed438164663d9be","reason":"Same source body in authored semantic page"},{"oldId":"b-f333d068c199c798f5e050d1","targetId":"b-f333d068c199c798f5e050d1","page":"reading-opening","operation":"move","beforeSHA256":"c69ff8d382b78b38f116e433914d074c2acb001bb346dd6ea11428f503fb1e9b","afterSHA256":"c69ff8d382b78b38f116e433914d074c2acb001bb346dd6ea11428f503fb1e9b","reason":"Same source body in authored semantic page"},{"oldId":"b-778588b51b6fcd36ef46812d","targetId":"b-778588b51b6fcd36ef46812d","page":"reading-opening","operation":"move","beforeSHA256":"ca0781f38ff1958fb130064abb90ad8c2678b32c1241d77839edd7ae1ef6acf8","afterSHA256":"ca0781f38ff1958fb130064abb90ad8c2678b32c1241d77839edd7ae1ef6acf8","reason":"Same source body in authored semantic page"},{"oldId":"b-a3580745ef2bffbe51fbb5cf","targetId":"b-a3580745ef2bffbe51fbb5cf","page":"reading-opening","operation":"move","beforeSHA256":"2263210ab03d3652966597d08ee87e2d2abef3bd071157eb51a6ea67ca8520ea","afterSHA256":"2263210ab03d3652966597d08ee87e2d2abef3bd071157eb51a6ea67ca8520ea","reason":"Same source body in authored semantic page"},{"oldId":"b-ce949d2c054572fcfb505fcc","targetId":"b-ce949d2c054572fcfb505fcc","page":"reading-opening","operation":"move","beforeSHA256":"92f197c441dd9a60ae9246f7eb4d102d3d469d13a8a8de0407e0c9c9aa009e0d","afterSHA256":"92f197c441dd9a60ae9246f7eb4d102d3d469d13a8a8de0407e0c9c9aa009e0d","reason":"Same source body in authored semantic page"},{"oldId":"b-0e7504e6e3acf53fe05f351f","targetId":"b-0e7504e6e3acf53fe05f351f","page":"reading-opening","operation":"move","beforeSHA256":"2b9ccef7e1220e2250cceff1abd8bdd736b42f8da5a223f26b029f44b962e807","afterSHA256":"2b9ccef7e1220e2250cceff1abd8bdd736b42f8da5a223f26b029f44b962e807","reason":"Same source body in authored semantic page"}],"blocks":[{"id":"b-5336bea76536cfaddfdb696c","type":"h1","scope":"caption-decoding-later","url":"https://codingai-lec03.pages.dev/course/blocks/caption-decoding-later.html#b-5336bea76536cfaddfdb696c","content":"# 选读：局部最高概率为何不保证整句最高概率","sha256":"07ef2b2a2dc1aab1aacdb60c913d70746a882f51b820c051f18b7d1c30021afd","chapter":{"id":"III","title":"视觉与语言"},"unit":{"id":"generation-many-possible-results","title":"一个条件怎样容纳多种合理结果？"},"page":{"id":"reading-opening","title":"选读：局部最高概率为何不保证整句最高概率","number":1},"sectionTitle":"选读：局部最高概率为何不保证整句最高概率","currentQuestion":"选读：局部最高概率为何不保证整句最高概率","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":null,"nextBlock":"b-289caffc80977c5811065725","prerequisites":[{"id":"b-289caffc80977c5811065725","url":"#b-289caffc80977c5811065725","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-289caffc80977c5811065725","type":"p","scope":"caption-decoding-later","url":"https://codingai-lec03.pages.dev/course/blocks/caption-decoding-later.html#b-289caffc80977c5811065725","content":"后续课程预备或按需选读，不列为本讲必讲内容。","sha256":"513b4dde07c19aa85a702ba5af2c592cb8b134338c12e49e6c6f6faf9adf92ed","chapter":{"id":"III","title":"视觉与语言"},"unit":{"id":"generation-many-possible-results","title":"一个条件怎样容纳多种合理结果？"},"page":{"id":"reading-opening","title":"选读：局部最高概率为何不保证整句最高概率","number":1},"sectionTitle":"选读：局部最高概率为何不保证整句最高概率","currentQuestion":"选读：局部最高概率为何不保证整句最高概率","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-5336bea76536cfaddfdb696c","nextBlock":"b-6a09f2304f5fd641979b4394","prerequisites":[{"id":"b-5336bea76536cfaddfdb696c","url":"#b-5336bea76536cfaddfdb696c","role":"section-title"}],"relatedBlocks":[]},{"id":"b-6a09f2304f5fd641979b4394","type":"p","scope":"caption-decoding-later","url":"https://codingai-lec03.pages.dev/course/blocks/caption-decoding-later.html#b-6a09f2304f5fd641979b4394","content":"先沿画面指出视觉特征进入哪里，再指出已有文字与下一词目标。连续说明展开 shifted target、EOS、PAD 和实际生成，LAB13 提供已保存的实验结果。理解本页可先把文字预测器作为输入输出模块；整数移位只演示目标构造，完整图像描述还需要图像条件与实际生成。","sha256":"aeaad29331bc2c94ac240ece646024336e7455de08d754b00ed438164663d9be","chapter":{"id":"III","title":"视觉与语言"},"unit":{"id":"generation-many-possible-results","title":"一个条件怎样容纳多种合理结果？"},"page":{"id":"reading-opening","title":"选读：局部最高概率为何不保证整句最高概率","number":1},"sectionTitle":"选读：局部最高概率为何不保证整句最高概率","currentQuestion":"选读：局部最高概率为何不保证整句最高概率","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-289caffc80977c5811065725","nextBlock":"b-f333d068c199c798f5e050d1","prerequisites":[{"id":"b-5336bea76536cfaddfdb696c","url":"#b-5336bea76536cfaddfdb696c","role":"section-title"},{"id":"b-289caffc80977c5811065725","url":"#b-289caffc80977c5811065725","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-f333d068c199c798f5e050d1","type":"p","scope":"caption-decoding-later","url":"https://codingai-lec03.pages.dev/course/blocks/caption-decoding-later.html#b-f333d068c199c798f5e050d1","content":"这段作为第四讲生成和第七讲搜索的预备阅读；第三讲只需知道推理会接续自己的输出。令首词 A/B 概率为 0.6/0.4；A 后 X/Y 为 0.51/0.49；B 后 X/Y 为 0.9/0.1；下一步都以概率 1 输出 EOS。每层出边和为 1。","sha256":"c69ff8d382b78b38f116e433914d074c2acb001bb346dd6ea11428f503fb1e9b","chapter":{"id":"III","title":"视觉与语言"},"unit":{"id":"generation-many-possible-results","title":"一个条件怎样容纳多种合理结果？"},"page":{"id":"reading-opening","title":"选读：局部最高概率为何不保证整句最高概率","number":1},"sectionTitle":"选读：局部最高概率为何不保证整句最高概率","currentQuestion":"选读：局部最高概率为何不保证整句最高概率","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-6a09f2304f5fd641979b4394","nextBlock":"b-778588b51b6fcd36ef46812d","prerequisites":[{"id":"b-5336bea76536cfaddfdb696c","url":"#b-5336bea76536cfaddfdb696c","role":"section-title"},{"id":"b-289caffc80977c5811065725","url":"#b-289caffc80977c5811065725","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-778588b51b6fcd36ef46812d","type":"p","scope":"caption-decoding-later","url":"https://codingai-lec03.pages.dev/course/blocks/caption-decoding-later.html#b-778588b51b6fcd36ef46812d","content":"完整路径 AX、AY、BX、BY 的概率分别为 0.306、0.294、0.360、0.040。greedy 首先取 A，再取 X，得到 0.306；宽度 2 的 beam 首步保留 A/B，再展开并排序，能保留最终的 BX=0.36。这个可算小树解释局部选择与完整路径概率的差别。","sha256":"ca0781f38ff1958fb130064abb90ad8c2678b32c1241d77839edd7ae1ef6acf8","chapter":{"id":"III","title":"视觉与语言"},"unit":{"id":"generation-many-possible-results","title":"一个条件怎样容纳多种合理结果？"},"page":{"id":"reading-opening","title":"选读：局部最高概率为何不保证整句最高概率","number":1},"sectionTitle":"选读：局部最高概率为何不保证整句最高概率","currentQuestion":"选读：局部最高概率为何不保证整句最高概率","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-f333d068c199c798f5e050d1","nextBlock":"b-a3580745ef2bffbe51fbb5cf","prerequisites":[{"id":"b-5336bea76536cfaddfdb696c","url":"#b-5336bea76536cfaddfdb696c","role":"section-title"},{"id":"b-289caffc80977c5811065725","url":"#b-289caffc80977c5811065725","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-a3580745ef2bffbe51fbb5cf","type":"p","scope":"caption-decoding-later","url":"https://codingai-lec03.pages.dev/course/blocks/caption-decoding-later.html#b-a3580745ef2bffbe51fbb5cf","content":"有限 beam 仍可能在更大树里提前删掉最优前缀；更高模型概率也不保证更符合图像。如果模型把蓝杯误识成红杯，搜索更充分可能只是更坚定地输出错误描述。必须区分视觉建模错误、概率预测错误和搜索错误。","sha256":"2263210ab03d3652966597d08ee87e2d2abef3bd071157eb51a6ea67ca8520ea","chapter":{"id":"III","title":"视觉与语言"},"unit":{"id":"generation-many-possible-results","title":"一个条件怎样容纳多种合理结果？"},"page":{"id":"reading-opening","title":"选读：局部最高概率为何不保证整句最高概率","number":1},"sectionTitle":"选读：局部最高概率为何不保证整句最高概率","currentQuestion":"选读：局部最高概率为何不保证整句最高概率","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-778588b51b6fcd36ef46812d","nextBlock":"b-ce949d2c054572fcfb505fcc","prerequisites":[{"id":"b-5336bea76536cfaddfdb696c","url":"#b-5336bea76536cfaddfdb696c","role":"section-title"},{"id":"b-289caffc80977c5811065725","url":"#b-289caffc80977c5811065725","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-ce949d2c054572fcfb505fcc","type":"p","scope":"caption-decoding-later","url":"https://codingai-lec03.pages.dev/course/blocks/caption-decoding-later.html#b-ce949d2c054572fcfb505fcc","content":"当前 LAB13 的实际模型中 greedy 与 beam=3 均为 100% 整句匹配，没有显示 beam 的经验优势。其热身代码采用另一棵归一化概率树，greedy=AZ、概率 0.3025，beam=BY、概率 0.36；两棵树各自定义自洽，不能把一棵树的首步概率与另一棵树的结果混用。","sha256":"92f197c441dd9a60ae9246f7eb4d102d3d469d13a8a8de0407e0c9c9aa009e0d","chapter":{"id":"III","title":"视觉与语言"},"unit":{"id":"generation-many-possible-results","title":"一个条件怎样容纳多种合理结果？"},"page":{"id":"reading-opening","title":"选读：局部最高概率为何不保证整句最高概率","number":1},"sectionTitle":"选读：局部最高概率为何不保证整句最高概率","currentQuestion":"选读：局部最高概率为何不保证整句最高概率","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-a3580745ef2bffbe51fbb5cf","nextBlock":"b-0e7504e6e3acf53fe05f351f","prerequisites":[{"id":"b-5336bea76536cfaddfdb696c","url":"#b-5336bea76536cfaddfdb696c","role":"section-title"},{"id":"b-289caffc80977c5811065725","url":"#b-289caffc80977c5811065725","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-0e7504e6e3acf53fe05f351f","type":"h2","scope":"caption-decoding-later","url":"https://codingai-lec03.pages.dev/course/blocks/caption-decoding-later.html#b-0e7504e6e3acf53fe05f351f","content":"## 继续阅读与实验","sha256":"2b9ccef7e1220e2250cceff1abd8bdd736b42f8da5a223f26b029f44b962e807","chapter":{"id":"III","title":"视觉与语言"},"unit":{"id":"generation-many-possible-results","title":"一个条件怎样容纳多种合理结果？"},"page":{"id":"reading-opening","title":"选读：局部最高概率为何不保证整句最高概率","number":1},"sectionTitle":"选读：局部最高概率为何不保证整句最高概率","currentQuestion":"选读：局部最高概率为何不保证整句最高概率","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-ce949d2c054572fcfb505fcc","nextBlock":null,"prerequisites":[{"id":"b-5336bea76536cfaddfdb696c","url":"#b-5336bea76536cfaddfdb696c","role":"section-title"},{"id":"b-289caffc80977c5811065725","url":"#b-289caffc80977c5811065725","role":"problem-context"}],"relatedBlocks":[]}]}