{"version":2,"relationshipURLBase":"https://codingai-lec03.pages.dev/course/blocks/vl-history-convergence.html","relationshipResolution":"Resolve relative relationship URLs against relationshipURLBase; complete records and absolute URLs are in blocks.","title":"2014—2016：组件到位后，新的研究问题怎样出现","url":"https://codingai-lec03.pages.dev/course/blocks/vl-history-convergence.html","markdown":"https://codingai-lec03.pages.dev/course/blocks/vl-history-convergence.md","pages":[{"id":"reading-opening","title":"2014—2016：组件到位后，新的研究问题怎样出现","unit":"语言加入后，任务究竟改变了什么？","unitId":"language-task-interfaces","chapter":"III","number":1}],"aliases":[{"oldId":"b-802d395ef0345dcc0df62a10","targetId":"b-802d395ef0345dcc0df62a10","page":"reading-opening","operation":"move","beforeSHA256":"c6f3930653c3b44fc42e74dbb2f6a02ca8b2e94970b61b00a96fb475344c2e72","afterSHA256":"c6f3930653c3b44fc42e74dbb2f6a02ca8b2e94970b61b00a96fb475344c2e72","reason":"Same source body in authored semantic page"},{"oldId":"b-65a58e7118156719fad6d88c","targetId":"b-65a58e7118156719fad6d88c","page":"reading-opening","operation":"move","beforeSHA256":"7ea95f3bede47f014c7bc900fc9a2b5a7450de2656ad4555a6b44874a7f1a91e","afterSHA256":"7ea95f3bede47f014c7bc900fc9a2b5a7450de2656ad4555a6b44874a7f1a91e","reason":"Same source body in authored semantic page"},{"oldId":"b-08d481a3b77ba7537f53cf8a","targetId":"b-08d481a3b77ba7537f53cf8a","page":"reading-opening","operation":"move","beforeSHA256":"e38ac802ecd090b6e18a5381cb2b8fba5151c603bcc32b461b93af6b5e090784","afterSHA256":"e38ac802ecd090b6e18a5381cb2b8fba5151c603bcc32b461b93af6b5e090784","reason":"Same source body in authored semantic page"},{"oldId":"b-0c98058af1dc70549c854142","targetId":"b-0c98058af1dc70549c854142","page":"reading-opening","operation":"move","beforeSHA256":"6100e7cba5016da14069f3fe3f0d465d4ac2630cda48b1a9a0172f753d2599d0","afterSHA256":"6100e7cba5016da14069f3fe3f0d465d4ac2630cda48b1a9a0172f753d2599d0","reason":"Same source body in authored semantic page"},{"oldId":"b-0ff8b7ed0bc432f3f1e7819f","targetId":"b-0ff8b7ed0bc432f3f1e7819f","page":"reading-opening","operation":"move","beforeSHA256":"3fc789ae4a4fc25185bc795b7579ec7f9f700288873cf3509808f033e1a79263","afterSHA256":"3fc789ae4a4fc25185bc795b7579ec7f9f700288873cf3509808f033e1a79263","reason":"Same source body in authored semantic page"},{"oldId":"b-93bc9b2b9ab8132d3dbf5394","targetId":"b-93bc9b2b9ab8132d3dbf5394","page":"reading-opening","operation":"move","beforeSHA256":"76cc3f85a5998404c19a2107d834cc1d9137e119b90caec54278bd7856f66d6b","afterSHA256":"76cc3f85a5998404c19a2107d834cc1d9137e119b90caec54278bd7856f66d6b","reason":"Same source body in authored semantic page"},{"oldId":"b-74631fdcb6f9dc631c012c56","targetId":"b-74631fdcb6f9dc631c012c56","page":"reading-opening","operation":"rewrite","beforeSHA256":"f73dd32177ce36a093e27fd687dda5fbf823a8ce109377dea17f58ec92da6f8e","afterSHA256":"151d7b56836661552ee9a7d90c2120f585c62443f088e515d60a7ecdb15ba8ae","reason":"Same source identity with figure semantics or reading context enriched"},{"oldId":"b-ab0ab7f60ee8896ffa088502","targetId":"b-ab0ab7f60ee8896ffa088502","page":"reading-opening","operation":"move","beforeSHA256":"2b9ccef7e1220e2250cceff1abd8bdd736b42f8da5a223f26b029f44b962e807","afterSHA256":"2b9ccef7e1220e2250cceff1abd8bdd736b42f8da5a223f26b029f44b962e807","reason":"Same source body in authored semantic page"}],"blocks":[{"id":"b-802d395ef0345dcc0df62a10","type":"h1","scope":"vl-history-convergence","url":"https://codingai-lec03.pages.dev/course/blocks/vl-history-convergence.html#b-802d395ef0345dcc0df62a10","content":"# 2014—2016：组件到位后，新的研究问题怎样出现","sha256":"c6f3930653c3b44fc42e74dbb2f6a02ca8b2e94970b61b00a96fb475344c2e72","chapter":{"id":"III","title":"视觉与语言"},"unit":{"id":"language-task-interfaces","title":"语言加入后，任务究竟改变了什么？"},"page":{"id":"reading-opening","title":"2014—2016：组件到位后，新的研究问题怎样出现","number":1},"sectionTitle":"2014—2016：组件到位后，新的研究问题怎样出现","currentQuestion":"2014—2016：组件到位后，新的研究问题怎样出现","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":null,"nextBlock":"b-65a58e7118156719fad6d88c","prerequisites":[{"id":"b-65a58e7118156719fad6d88c","url":"#b-65a58e7118156719fad6d88c","role":"problem-context"}],"relatedBlocks":[{"id":"b-74631fdcb6f9dc631c012c56","url":"#b-74631fdcb6f9dc631c012c56","relation":"figure-and-explanation"}]},{"id":"b-65a58e7118156719fad6d88c","type":"p","scope":"vl-history-convergence","url":"https://codingai-lec03.pages.dev/course/blocks/vl-history-convergence.html#b-65a58e7118156719fad6d88c","content":"本页把研究史当作问题史阅读：视觉表示、文字预测和数据已经可用，研究者分别选择描述、局部对应与问答。对照这些工作各自改变了什么、取得了什么结果，再理解研究者为什么继续推进。公开人物背景用于定位当时的研究环境。","sha256":"7ea95f3bede47f014c7bc900fc9a2b5a7450de2656ad4555a6b44874a7f1a91e","chapter":{"id":"III","title":"视觉与语言"},"unit":{"id":"language-task-interfaces","title":"语言加入后，任务究竟改变了什么？"},"page":{"id":"reading-opening","title":"2014—2016：组件到位后，新的研究问题怎样出现","number":1},"sectionTitle":"2014—2016：组件到位后，新的研究问题怎样出现","currentQuestion":"2014—2016：组件到位后，新的研究问题怎样出现","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-802d395ef0345dcc0df62a10","nextBlock":"b-08d481a3b77ba7537f53cf8a","prerequisites":[{"id":"b-802d395ef0345dcc0df62a10","url":"#b-802d395ef0345dcc0df62a10","role":"section-title"}],"relatedBlocks":[{"id":"b-74631fdcb6f9dc631c012c56","url":"#b-74631fdcb6f9dc631c012c56","relation":"figure-and-explanation"}]},{"id":"b-08d481a3b77ba7537f53cf8a","type":"p","scope":"vl-history-convergence","url":"https://codingai-lec03.pages.dev/course/blocks/vl-history-convergence.html#b-08d481a3b77ba7537f53cf8a","content":"这段历史可以按“哪项能力已经可复用，哪条关系还没有学会”阅读。卷积网络已经能提供视觉特征；文字预测器能组织可变长度输出；图文配对和共同评价数据使不同系统能在同一问题上比较。于是连接视觉与语言开始具有可操作的实验入口。","sha256":"e38ac802ecd090b6e18a5381cb2b8fba5151c603bcc32b461b93af6b5e090784","chapter":{"id":"III","title":"视觉与语言"},"unit":{"id":"language-task-interfaces","title":"语言加入后，任务究竟改变了什么？"},"page":{"id":"reading-opening","title":"2014—2016：组件到位后，新的研究问题怎样出现","number":1},"sectionTitle":"2014—2016：组件到位后，新的研究问题怎样出现","currentQuestion":"2014—2016：组件到位后，新的研究问题怎样出现","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-65a58e7118156719fad6d88c","nextBlock":"b-0c98058af1dc70549c854142","prerequisites":[{"id":"b-802d395ef0345dcc0df62a10","url":"#b-802d395ef0345dcc0df62a10","role":"section-title"},{"id":"b-65a58e7118156719fad6d88c","url":"#b-65a58e7118156719fad6d88c","role":"problem-context"}],"relatedBlocks":[{"id":"b-74631fdcb6f9dc631c012c56","url":"#b-74631fdcb6f9dc631c012c56","relation":"figure-and-explanation"}]},{"id":"b-0c98058af1dc70549c854142","type":"p","scope":"vl-history-convergence","url":"https://codingai-lec03.pages.dev/course/blocks/vl-history-convergence.html#b-0c98058af1dc70549c854142","content":"LRCN 和 Show and Tell 在 2014 年发布预印本、2015 年发表，研究怎样把视觉信息送入文字输出。区域—句子对齐把对应关系细化到局部；VQA 则让问题决定当前要读什么。它们不是同一系统的连续版本，不能画成“描述完成后才有对齐，再完成后才有问答”的单行升级史。","sha256":"6100e7cba5016da14069f3fe3f0d465d4ac2630cda48b1a9a0172f753d2599d0","chapter":{"id":"III","title":"视觉与语言"},"unit":{"id":"language-task-interfaces","title":"语言加入后，任务究竟改变了什么？"},"page":{"id":"reading-opening","title":"2014—2016：组件到位后，新的研究问题怎样出现","number":1},"sectionTitle":"2014—2016：组件到位后，新的研究问题怎样出现","currentQuestion":"2014—2016：组件到位后，新的研究问题怎样出现","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-08d481a3b77ba7537f53cf8a","nextBlock":"b-0ff8b7ed0bc432f3f1e7819f","prerequisites":[{"id":"b-802d395ef0345dcc0df62a10","url":"#b-802d395ef0345dcc0df62a10","role":"section-title"},{"id":"b-65a58e7118156719fad6d88c","url":"#b-65a58e7118156719fad6d88c","role":"problem-context"}],"relatedBlocks":[{"id":"b-74631fdcb6f9dc631c012c56","url":"#b-74631fdcb6f9dc631c012c56","relation":"figure-and-explanation"}]},{"id":"b-0ff8b7ed0bc432f3f1e7819f","type":"p","scope":"vl-history-convergence","url":"https://codingai-lec03.pages.dev/course/blocks/vl-history-convergence.html#b-0ff8b7ed0bc432f3f1e7819f","content":"研究进展不仅是多了一种漂亮演示。Show and Tell 在其论文中的 COCO 协议报告 BLEU-4 27.7，并在多个数据集比较描述质量；这个历史结果说明学习图像条件的句子分布能够产出可比较的描述系统。数字属于当时模型与评价协议，不是本课程 Lab 的分数。区域对齐工作又在图文检索和区域描述中展示了局部对应的价值。这里“为什么值得继续研究”的答案，是新接口已经带来可测量的任务能力，随后才能进一步问它遗漏了什么。","sha256":"3fc789ae4a4fc25185bc795b7579ec7f9f700288873cf3509808f033e1a79263","chapter":{"id":"III","title":"视觉与语言"},"unit":{"id":"language-task-interfaces","title":"语言加入后，任务究竟改变了什么？"},"page":{"id":"reading-opening","title":"2014—2016：组件到位后，新的研究问题怎样出现","number":1},"sectionTitle":"2014—2016：组件到位后，新的研究问题怎样出现","currentQuestion":"2014—2016：组件到位后，新的研究问题怎样出现","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-0c98058af1dc70549c854142","nextBlock":"b-93bc9b2b9ab8132d3dbf5394","prerequisites":[{"id":"b-802d395ef0345dcc0df62a10","url":"#b-802d395ef0345dcc0df62a10","role":"section-title"},{"id":"b-65a58e7118156719fad6d88c","url":"#b-65a58e7118156719fad6d88c","role":"problem-context"}],"relatedBlocks":[{"id":"b-74631fdcb6f9dc631c012c56","url":"#b-74631fdcb6f9dc631c012c56","relation":"figure-and-explanation"}]},{"id":"b-93bc9b2b9ab8132d3dbf5394","type":"p","scope":"vl-history-convergence","url":"https://codingai-lec03.pages.dev/course/blocks/vl-history-convergence.html#b-93bc9b2b9ab8132d3dbf5394","content":"Trevor Darrell 是 LRCN 等工作的共同作者，其公开研究涉及视觉感知与语言、人机交互的连接。课程借此展示一条公共研究脉络；个人经历、合作动机或具体对话不由论文作者名单推断。","sha256":"76cc3f85a5998404c19a2107d834cc1d9137e119b90caec54278bd7856f66d6b","chapter":{"id":"III","title":"视觉与语言"},"unit":{"id":"language-task-interfaces","title":"语言加入后，任务究竟改变了什么？"},"page":{"id":"reading-opening","title":"2014—2016：组件到位后，新的研究问题怎样出现","number":1},"sectionTitle":"2014—2016：组件到位后，新的研究问题怎样出现","currentQuestion":"2014—2016：组件到位后，新的研究问题怎样出现","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-0ff8b7ed0bc432f3f1e7819f","nextBlock":"b-74631fdcb6f9dc631c012c56","prerequisites":[{"id":"b-802d395ef0345dcc0df62a10","url":"#b-802d395ef0345dcc0df62a10","role":"section-title"},{"id":"b-65a58e7118156719fad6d88c","url":"#b-65a58e7118156719fad6d88c","role":"problem-context"}],"relatedBlocks":[{"id":"b-74631fdcb6f9dc631c012c56","url":"#b-74631fdcb6f9dc631c012c56","relation":"figure-and-explanation"}]},{"id":"b-74631fdcb6f9dc631c012c56","type":"figure","scope":"vl-history-convergence","url":"https://codingai-lec03.pages.dev/course/blocks/vl-history-convergence.html#b-74631fdcb6f9dc631c012c56","content":"图内标签与数值：可复用视觉特征；文字序列预测；图文数据与共同评价；分类预训练；区域表示；已有文字 → 后续文字；可变长度输出；图像—描述；图像—问题—答案；2013；DeViSE；视觉—语义空间；2014；LRCN / Show and Tell；图像描述；2015；区域—句子对齐 / VQA；局部对应与问题条件；2016；DCC；未配对类别的描述；Trevor Darrell；MIT 教职 1999—2008 → Berkeley，2008；R-CNN、FCN、LRCN 的共同作者之一；积累汇合，让新问题可训练：；说出内容 · 对应局部 · 按问题回答；年份标示预印本或会议窗口；技术联系不等于私人动机或单一研究者的谱系。\n\n时间带展示几条研究路线在相近时期汇合，不表示单一方法的替代顺序。 查看对应图解状态\n\n- 可复用视觉特征\n- 文字序列预测\n- 图文数据与共同评价\n- 分类预训练\n- 区域表示\n- 已有文字 → 后续文字\n- 可变长度输出\n- 图像—描述\n- 图像—问题—答案\n- 2013\n- DeViSE\n- 视觉—语义空间\n- 2014\n- LRCN / Show and Tell\n- 图像描述\n- 2015\n- 区域—句子对齐 / VQA\n- 局部对应与问题条件\n- 2016\n- DCC\n- 未配对类别的描述\n- Trevor Darrell\n- MIT 教职 1999—2008 → Berkeley，2008\n- R-CNN、FCN、LRCN 的共同作者之一\n- 积累汇合，让新问题可训练：\n- 说出内容 · 对应局部 · 按问题回答\n- 年份标示预印本或会议窗口；技术联系不等于私人动机或单一研究者的谱系。时间带展示几条研究路线在相近时期汇合，不表示单一方法的替代顺序。 [查看对应图解状态](/#/l3-r2-p50/0)[查看完整图解](/#/l3-r2-p50/0)","sha256":"151d7b56836661552ee9a7d90c2120f585c62443f088e515d60a7ecdb15ba8ae","chapter":{"id":"III","title":"视觉与语言"},"unit":{"id":"language-task-interfaces","title":"语言加入后，任务究竟改变了什么？"},"page":{"id":"reading-opening","title":"2014—2016：组件到位后，新的研究问题怎样出现","number":1},"sectionTitle":"2014—2016：组件到位后，新的研究问题怎样出现","currentQuestion":"2014—2016：组件到位后，新的研究问题怎样出现","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"figure":{"caption":"时间带展示几条研究路线在相近时期汇合，不表示单一方法的替代顺序。 查看对应图解状态","labels":["可复用视觉特征","文字序列预测","图文数据与共同评价","分类预训练","区域表示","已有文字 → 后续文字","可变长度输出","图像—描述","图像—问题—答案","2013","DeViSE","视觉—语义空间","2014","LRCN / Show and Tell","图像描述","2015","区域—句子对齐 / VQA","局部对应与问题条件","2016","DCC","未配对类别的描述","Trevor Darrell","MIT 教职 1999—2008 → Berkeley，2008","R-CNN、FCN、LRCN 的共同作者之一","积累汇合，让新问题可训练：","说出内容 · 对应局部 · 按问题回答","年份标示预印本或会议窗口；技术联系不等于私人动机或单一研究者的谱系。"],"description":"时间带展示几条研究路线在相近时期汇合，不表示单一方法的替代顺序。 查看对应图解状态可复用视觉特征文字序列预测图文数据与共同评价分类预训练区域表示已有文字 → 后续文字可变长度输出图像—描述图像—问题—答案2013DeViSE视觉—语义空间2014LRCN / Show and Tell图像描述2015区域—句子对齐 / VQA局部对应与问题条件2016DCC未配对类别的描述Trevor DarrellMIT 教职 1999—2008 → Berkeley，2008R-CNN、FCN、LRCN 的共同作者之一积累汇合，让新问题可训练：说出内容 · 对应局部 · 按问题回答年份标示预印本或会议窗口；技术联系不等于私人动机或单一研究者的谱系。","evidenceIdentity":"时间带展示几条研究路线在相近时期汇合，不表示单一方法的替代顺序。 查看对应图解状态","images":[]},"previousBlock":"b-93bc9b2b9ab8132d3dbf5394","nextBlock":"b-ab0ab7f60ee8896ffa088502","prerequisites":[{"id":"b-802d395ef0345dcc0df62a10","url":"#b-802d395ef0345dcc0df62a10","role":"section-title"},{"id":"b-65a58e7118156719fad6d88c","url":"#b-65a58e7118156719fad6d88c","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-ab0ab7f60ee8896ffa088502","type":"h2","scope":"vl-history-convergence","url":"https://codingai-lec03.pages.dev/course/blocks/vl-history-convergence.html#b-ab0ab7f60ee8896ffa088502","content":"## 继续阅读与实验","sha256":"2b9ccef7e1220e2250cceff1abd8bdd736b42f8da5a223f26b029f44b962e807","chapter":{"id":"III","title":"视觉与语言"},"unit":{"id":"language-task-interfaces","title":"语言加入后，任务究竟改变了什么？"},"page":{"id":"reading-opening","title":"2014—2016：组件到位后，新的研究问题怎样出现","number":1},"sectionTitle":"2014—2016：组件到位后，新的研究问题怎样出现","currentQuestion":"2014—2016：组件到位后，新的研究问题怎样出现","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-74631fdcb6f9dc631c012c56","nextBlock":null,"prerequisites":[{"id":"b-802d395ef0345dcc0df62a10","url":"#b-802d395ef0345dcc0df62a10","role":"section-title"},{"id":"b-65a58e7118156719fad6d88c","url":"#b-65a58e7118156719fad6d88c","role":"problem-context"}],"relatedBlocks":[{"id":"b-74631fdcb6f9dc631c012c56","url":"#b-74631fdcb6f9dc631c012c56","relation":"figure-and-explanation"}]}]}
