[{"data":1,"prerenderedAt":1556},["ShallowReactive",2],{"blog-taxonomies":3,"blog-stats":507,"post-2026\u002F2026-01-06-人工智能导论":1083},[4,11,16,21,25,29,36,39,44,47,52,58,63,66,71,76,80,84,87,93,96,101,105,110,113,117,122,127,132,136,141,143,146,151,155,157,162,165,169,173,176,179,183,186,190,193,197,200,205,208,211,214,219,222,228,234,237,239,245,249,253,256,260,266,271,276,281,283,287,290,293,298,301,304,306,308,311,314,316,318,321,327,332,337,342,347,352,354,356,358,361,365,367,369,371,374,376,378,381,386,390,393,396,400,404,407,411,413,415,417,421,423,425,428,431,433,435,437,439,441,443,446,448,450,454,456,459,461,463,465,467,469,471,473,477,483,486,488,491,493,495,497,499,501,503,505],{"category":5,"tags":6},"笔记",[7,8,9,10],"QQ","插件","LiteLoader","美化",{"category":12,"tags":13},"其他",[14,15,10],"博客","IP签名",{"category":17,"tags":18},"杂谈",[19,20],"互联网","观点",{"category":5,"tags":22},[14,23,24],"GitHub","数据可视化",{"category":5,"tags":26},[14,27,28],"访问统计","不蒜子",{"category":30,"tags":31},"工具",[32,33,34,35],"Python","爬虫","选课脚本","广州大学",{"category":5,"tags":37},[38,5],"数学分析",{"category":40,"tags":41},"算法",[42,40,43],"数据结构","线性表",{"category":40,"tags":45},[42,46],"复习",{"category":5,"tags":48},[49,50,51],"高等代数","线性代数","知识点",{"category":5,"tags":53},[14,54,55,56,57],"Typora","PicGo","图床","Markdown",{"category":30,"tags":59},[60,61,62],"Jupyter","工具配置","中文",{"category":30,"tags":64},[65,61,10],"VSCode",{"category":30,"tags":67},[68,69,70,35],"路由器","OpenWrt","校园网",{"category":12,"tags":72},[73,74,75],"网络协议","Wireshark","DEVP2P",{"category":5,"tags":77},[73,74,78,79],"Lua","解码器",{"category":5,"tags":81},[82,5,83],"常微分方程","数学",{"category":5,"tags":85},[86,5,83],"概率论",{"category":30,"tags":88},[89,90,91,92],"Docker","WSL","Windows","磁盘清理",{"category":30,"tags":94},[32,33,95,35],"课程表",{"category":5,"tags":97},[14,98,99,100],"Hexo","Gitalk","评论系统",{"category":5,"tags":102},[14,103,98,104],"Obsidian","写作工具",{"category":12,"tags":106},[107,108,109],"课题研究","泊松分布","参数估计",{"category":12,"tags":111},[107,108,109,112],"组会",{"category":12,"tags":114},[109,115,108,116],"IM模型","统计学",{"category":17,"tags":118},[119,120,121],"软考","网络规划","信息系统项目管理",{"category":30,"tags":123},[124,125,126],"Arch Linux","AppImage","Linux",{"category":5,"tags":128},[129,130,131],"编程语言","计算机科学","内存管理",{"category":5,"tags":133},[134,135],"Canvas","前端",{"category":5,"tags":137},[14,138,139,140],"GitHub Actions","Vite","部署",{"category":5,"tags":142},[135],{"category":5,"tags":144},[135,145],"系统设计",{"category":5,"tags":147},[135,148,149,150],"安全","XSS","CSRF",{"category":152,"tags":153},"项目",[134,154],"PixiJS",{"category":152,"tags":156},null,{"category":40,"tags":158},[42,159,160,161],"栈","队列","JavaScript",{"category":40,"tags":163},[42,164,161],"哈希表",{"category":5,"tags":166},[167,161,168],"ES6","Set",{"category":40,"tags":170},[40,171,172],"动态规划","背包问题",{"category":40,"tags":174},[40,175],"滑动窗口",{"category":40,"tags":177},[42,178,161],"链表",{"category":40,"tags":180},[40,181,161,182],"ACM","面试",{"category":40,"tags":184},[40,161,185],"数据类型转换",{"category":40,"tags":187},[42,188,178,189],"顺序表","TypeScript",{"category":5,"tags":191},[135,154,134,192],"渲染",{"category":5,"tags":194},[135,195,134,196],"富文本编辑","wangEditor",{"category":40,"tags":198},[199,40,178],"力扣",{"category":5,"tags":201},[202,203,204,91],"DNS","加密","网络安全",{"category":5,"tags":206},[135,207],"Axios",{"category":5,"tags":209},[135,207,189,210],"接口封装",{"category":5,"tags":212},[135,213,145],"JWT",{"category":152,"tags":215},[134,216,217,154,218],"画布","React","项目开发",{"category":17,"tags":220},[221],"日志",{"category":5,"tags":223},[224,225,226,227],"Neo4j","图数据库","中心度分析","PageRank",{"category":12,"tags":229},[230,231,232,233],"论文","机器学习","随机森林","保险欺诈",{"category":5,"tags":235},[236],"人工智能",{"category":5,"tags":238},[231,40,32],{"category":12,"tags":240},[241,242,243,244],"共识算法","最长链","区块链","算法设计",{"category":5,"tags":246},[161,247,248,135],"闭包","作用域",{"category":5,"tags":250},[14,138,251,252],"RSS","自动化",{"category":30,"tags":254},[89,90,91,255],"安装",{"category":30,"tags":257},[258,91,259],"MSYS2","开发环境",{"category":5,"tags":261},[262,263,264,265],"团队协作","Git","Commit规范","代码审查",{"category":30,"tags":267},[268,269,270],"oh-my-posh","终端美化","CLI",{"category":5,"tags":272},[273,274,275],"上下文工程","Agent","RAG",{"category":5,"tags":277},[278,279,280],"前端工程化","技术选型","工程基建",{"category":5,"tags":282},[275],{"category":5,"tags":284},[285,286],"SDD","规格驱动开发",{"category":5,"tags":288},[289],"neovim",{"category":5,"tags":291},[292],"技术写作",{"category":5,"tags":294},[295,32,296,89,297],"FastAPI","后端","工程实践",{"category":152,"tags":299},[274,300],"Harness Engineering",{"category":5,"tags":302},[303],"RFC",{"category":152,"tags":305},[274,300,275],{"category":152,"tags":307},[274],{"category":40,"tags":309},[199,40,161,310],"两数之和",{"category":40,"tags":312},[199,40,161,313],"双指针",{"category":40,"tags":315},[199,40,161,178],{"category":40,"tags":317},[199,40,161,175],{"category":40,"tags":319},[199,40,161,320],"回溯",{"category":5,"tags":322},[323,324,325,326],"分布式处理与计算","Spark","Scala","RDD",{"category":5,"tags":328},[323,329,330,331],"线性回归","逻辑回归","监督学习",{"category":5,"tags":333},[323,334,335,336],"数值优化","牛顿法","收敛性",{"category":5,"tags":338},[323,339,340,341],"PCA","数据降维","正则化",{"category":5,"tags":343},[323,344,345,346],"聚类","层次聚类","K-Means",{"category":5,"tags":348},[323,349,350,351],"随机模拟","统计推断","EM算法",{"category":5,"tags":353},[24],{"category":5,"tags":355},[24],{"category":5,"tags":357},[24],{"category":40,"tags":359},[42,360],"图论",{"category":40,"tags":362},[42,363,364],"树","二叉树",{"category":40,"tags":366},[42,160],{"category":40,"tags":368},[42],{"category":40,"tags":370},[42,159],{"category":40,"tags":372},[42,40,373],"复杂度分析",{"category":40,"tags":375},[42,43,178],{"category":40,"tags":377},[42,43,188],{"category":40,"tags":379},[42,380],"基础概念",{"category":17,"tags":382},[383,384,385],"随记","照片","高中",{"category":17,"tags":387},[383,388,389],"团建","战地",{"category":17,"tags":391},[383,392],"生活",{"category":17,"tags":394},[383,395],"文学",{"category":17,"tags":397},[383,398,399],"SCP","模因",{"category":17,"tags":401},[383,402,403],"画集","艺术",{"category":17,"tags":405},[14,406,383],"写作",{"category":17,"tags":408},[383,409,410],"音乐","摇滚",{"category":17,"tags":412},[383,392],{"category":17,"tags":414},[383,409],{"category":17,"tags":416},[383,35,392],{"category":17,"tags":418},[383,419,420],"同人","手书",{"category":17,"tags":422},[383,392],{"category":17,"tags":424},[383,392],{"category":17,"tags":426},[383,427],"植物",{"category":17,"tags":429},[383,430],"读书",{"category":17,"tags":432},[383,392],{"category":17,"tags":434},[383,392],{"category":17,"tags":436},[383,392],{"category":17,"tags":438},[383,392],{"category":17,"tags":440},[383,392],{"category":17,"tags":442},[383,392],{"category":17,"tags":444},[14,445,383],"一周年",{"category":17,"tags":447},[383,392],{"category":17,"tags":449},[383,392],{"category":17,"tags":451},[35,452,453],"开源组织","SITE-193",{"category":17,"tags":455},[383,392],{"category":17,"tags":457},[383,458],"年末",{"category":17,"tags":460},[383,392],{"category":17,"tags":462},[383,392],{"category":17,"tags":464},[14,17],{"category":17,"tags":466},[383,392],{"category":17,"tags":468},[383,392],{"category":17,"tags":470},[383,392],{"category":17,"tags":472},[383,392],{"category":17,"tags":474},[383,475,476],"游戏","Control",{"category":30,"tags":478},[479,480,481,482],"Quadim","图像处理","四叉树","Rust",{"category":17,"tags":484},[14,485],"上线纪念",{"category":17,"tags":487},[383,395],{"category":30,"tags":489},[203,490],"测试",{"category":17,"tags":492},[383],{"category":17,"tags":494},[383,392],{"category":17,"tags":496},[383],{"category":5,"tags":498},[40],{"category":5,"tags":500},[40],{"category":5,"tags":502},[40],{"category":5,"tags":504},[40],{"category":5,"tags":506},[40],[508,512,516,520,524,528,532,536,540,544,548,552,556,560,564,568,572,576,579,583,587,590,594,598,602,606,610,614,618,622,626,630,634,638,642,646,650,654,658,662,666,670,674,678,682,686,690,694,698,702,706,710,714,718,722,726,730,734,738,742,746,750,754,758,762,766,770,774,778,782,786,790,794,798,801,805,809,813,817,821,825,829,833,837,841,845,849,853,857,861,865,869,873,877,881,885,889,893,897,901,904,907,911,915,919,923,927,930,934,938,942,946,950,954,957,961,965,969,973,977,981,984,988,992,996,1000,1004,1007,1011,1015,1019,1023,1027,1031,1035,1039,1043,1047,1051,1055,1059,1063,1067,1071,1075,1079],{"path":509,"words":510,"published":511,"date":156},"\u002Fposts\u002F2024\u002F2024-06-25-liteloaderqqnt",1296,"2024-06-25 11:03:08",{"path":513,"words":514,"published":515,"date":156},"\u002Fposts\u002F2024\u002F2024-06-26-ip签名",0,"2024-06-26 22:53:53",{"path":517,"words":518,"published":519,"date":156},"\u002Fposts\u002F2024\u002F2024-06-29-何加盐中文互联网正在加速崩塌",4853,"2024-06-29 00:47:13",{"path":521,"words":522,"published":523,"date":156},"\u002Fposts\u002F2024\u002F2024-07-12-github-chart",270,"2024-07-12 20:02:50",{"path":525,"words":526,"published":527,"date":156},"\u002Fposts\u002F2024\u002F2024-07-14-访客数统计",192,"2024-07-24 01:34:28",{"path":529,"words":530,"published":531,"date":156},"\u002Fposts\u002F2024\u002F2024-07-26-广大选课脚本",110,"2024-07-26 01:34:28",{"path":533,"words":534,"published":535,"date":156},"\u002Fposts\u002F2024\u002F2024-07-26-数学分析笔记其二",14,"2024-07-26 23:54:02",{"path":537,"words":538,"published":539,"date":156},"\u002Fposts\u002F2024\u002F2024-07-26-数据结构复习其二",12856,"2024-07-26 02:48:08",{"path":541,"words":542,"published":543,"date":156},"\u002Fposts\u002F2024\u002F2024-07-26-数据结构复习相关",32856,"2024-07-26 02:24:14",{"path":545,"words":546,"published":547,"date":156},"\u002Fposts\u002F2024\u002F2024-07-26-线性代数与空间解析几何知识点全汇总",6110,"2024-07-26 03:27:55",{"path":549,"words":550,"published":551,"date":156},"\u002Fposts\u002F2024\u002F2024-07-31-typora-picgo-兰空图床打造markdown写作环境",311,"2024-07-31 12:56:23",{"path":553,"words":554,"published":555,"date":156},"\u002Fposts\u002F2024\u002F2024-09-02-jupyter配置中文",441,"2024-09-02 23:49:05",{"path":557,"words":558,"published":559,"date":156},"\u002Fposts\u002F2024\u002F2024-09-02-自定义vscode背景图片",100,"2024-09-02 23:39:00",{"path":561,"words":562,"published":563,"date":156},"\u002Fposts\u002F2024\u002F2024-09-03-制作gzhu校园网路由器",1423,"2024-09-03 23:43:18",{"path":565,"words":566,"published":567,"date":156},"\u002Fposts\u002F2024\u002F2024-09-26-09-25-2024",136,"2024-09-26 00:49:26",{"path":569,"words":570,"published":571,"date":156},"\u002Fposts\u002F2024\u002F2024-11-04-创建一个网络包解码器分析devp2p协议",1943,"2024-11-04 16:45:19",{"path":573,"words":574,"published":575,"date":156},"\u002Fposts\u002F2024\u002F2025-03-09-2024一些笔记常微分",16,"2024-12-09 22:51:00",{"path":577,"words":574,"published":578,"date":156},"\u002Fposts\u002F2024\u002F2025-03-09-2024一些笔记概率论","2025-03-09 22:51:00",{"path":580,"words":581,"published":582,"date":156},"\u002Fposts\u002F2025\u002F2025-03-01-windows下释放docker占用的wsl空间",164,"2025-03-01 13:54:15",{"path":584,"words":585,"published":586,"date":156},"\u002Fposts\u002F2025\u002F2025-03-01-爬虫实战-爬取广州大学课程表",774,"2025-03-01 14:55:48",{"path":588,"words":530,"published":589,"date":156},"\u002Fposts\u002F2025\u002F2025-04-03-hexo集成gitalk的问题","2025-04-03 14:26:34",{"path":591,"words":592,"published":593,"date":156},"\u002Fposts\u002F2025\u002F2025-09-22-记一次配置obsidian配合hexo写博客",92,"2025-09-22 16:47:52",{"path":595,"words":596,"published":597,"date":156},"\u002Fposts\u002F2025\u002F2025-10-24-10-20课题",204,"2025-10-24 08:15:01",{"path":599,"words":600,"published":601,"date":156},"\u002Fposts\u002F2025\u002F2025-11-02-组会朝花夕拾",501,"2025-11-02 18:56:59",{"path":603,"words":604,"published":605,"date":156},"\u002Fposts\u002F2025\u002F2025-11-05-基于随机加权推断模型im的参数估计算法",481,"2025-11-05 20:51:41",{"path":607,"words":608,"published":609,"date":156},"\u002Fposts\u002F2025\u002F2025-11-09-关于软考高项网规",884,"2025-11-09 19:54:52",{"path":611,"words":612,"published":613,"date":156},"\u002Fposts\u002F2025\u002F2025-11-17-arch-linux运行appimage相关",365,"2025-11-17 09:59:59",{"path":615,"words":616,"published":617,"date":156},"\u002Fposts\u002F2025\u002F2025-11-19-相关概念",2058,"2025-11-19 11:06:20",{"path":619,"words":620,"published":621,"date":156},"\u002Fposts\u002F2025\u002F2025-11-21-web可视化实践canvas",6393,"2025-11-21 14:50:48",{"path":623,"words":624,"published":625,"date":156},"\u002Fposts\u002F2025\u002F2025-11-22-使用-github-actions-自动部署基于vite的项目到-github-pages",272,"2025-11-23 01:59:24",{"path":627,"words":628,"published":629,"date":156},"\u002Fposts\u002F2025\u002F2025-11-22-关于前端包管理器npmpnpmyarn和bun",4890,"2025-11-22 21:21:15",{"path":631,"words":632,"published":633,"date":156},"\u002Fposts\u002F2025\u002F2025-11-23-undo-redo-机制具体实现",7839,"2025-11-23 19:15:00",{"path":635,"words":636,"published":637,"date":156},"\u002Fposts\u002F2025\u002F2025-11-27-前端安全-关于xss与crsf",1866,"2025-11-28 02:05:25",{"path":639,"words":640,"published":641,"date":156},"\u002Fposts\u002F2025\u002F2025-12-05-现代协同-2d-画布编辑器pixijs-v8",4285,"2025-12-05 02:00:16",{"path":643,"words":644,"published":645,"date":156},"\u002Fposts\u002F2025\u002F2025-12-07-杂谈-课题组系统设计",3710,"2025-12-07 21:09:33",{"path":647,"words":648,"published":649,"date":156},"\u002Fposts\u002F2025\u002F2025-12-25-javascript中的数组方法与栈stack和队列queue的实现",474,"2025-11-29 16:58:15",{"path":651,"words":652,"published":653,"date":156},"\u002Fposts\u002F2025\u002F2025-12-25-关于javascript-实现哈希表",189,"2025-11-27 11:31:39",{"path":655,"words":656,"published":657,"date":156},"\u002Fposts\u002F2025\u002F2025-12-25-关于javascript的set方法",332,"2025-12-25 15:26:18",{"path":659,"words":660,"published":661,"date":156},"\u002Fposts\u002F2025\u002F2025-12-25-关于动态规划(背包问题为例)",1238,"2025-11-17 18:24:04",{"path":663,"words":664,"published":665,"date":156},"\u002Fposts\u002F2025\u002F2025-12-25-关于滑动窗口",712,"2025-11-20 15:04:40",{"path":667,"words":668,"published":669,"date":156},"\u002Fposts\u002F2025\u002F2025-12-25-关于链表(javascript)",500,"2025-11-25 13:00:48",{"path":671,"words":672,"published":673,"date":156},"\u002Fposts\u002F2025\u002F2025-12-25-面试算法acm模式构建构建输入输出模板",2051,"2025-12-25 23:09:45",{"path":675,"words":676,"published":677,"date":156},"\u002Fposts\u002F2025\u002F2025-12-26-javascript-数字数组字符串的处理",483,"2025-12-26 11:20:25",{"path":679,"words":680,"published":681,"date":156},"\u002Fposts\u002F2025\u002F2025-12-26-javascripttypescript-的顺序表链表实现",446,"2025-12-26 12:16:12",{"path":683,"words":684,"published":685,"date":156},"\u002Fposts\u002F2025\u002F2025-12-27-前端画布设计vol-1-实现基础yuan素渲染和状态控制",2136,"2025-12-28 01:23:08",{"path":687,"words":688,"published":689,"date":156},"\u002Fposts\u002F2025\u002F2025-12-27-前端画布设计vol-2-实现富文本编辑",1838,"2025-12-28 02:58:47",{"path":691,"words":692,"published":693,"date":156},"\u002Fposts\u002F2025\u002F2025-12-27-算法刷题-关于链表操作",2135,"2025-12-27 16:12:42",{"path":695,"words":696,"published":697,"date":156},"\u002Fposts\u002F2025\u002F2025-12-27-配置dnscrypt-proxy实现加密dns服务windows",1899,"2025-11-15 01:06:37",{"path":699,"words":700,"published":701,"date":156},"\u002Fposts\u002F2025\u002F2025-12-28-前端-关于网络请求xhrajaxfetchaxios",7788,"2025-10-28 15:48:46",{"path":703,"words":704,"published":705,"date":156},"\u002Fposts\u002F2025\u002F2025-12-28-前端-接口封装与请求规范axios为例",2373,"2025-12-29 01:01:18",{"path":707,"words":708,"published":709,"date":156},"\u002Fposts\u002F2025\u002F2025-12-28-前端-身份验证管理-基于-jwt-token-的实现",2709,"2025-11-23 22:04:41",{"path":711,"words":712,"published":713,"date":156},"\u002Fposts\u002F2025\u002F2025-12-29-记canvas画布项目开发",6254,"2025-12-29 04:52:15",{"path":715,"words":716,"published":717,"date":156},"\u002Fposts\u002F2025\u002Fabout-site",116,"2025-03-31 16:44:24",{"path":719,"words":720,"published":721,"date":156},"\u002Fposts\u002F2025\u002F使用neo4j图数据科学库gds进行中心度分析",143,"2025-04-02 23:37:24",{"path":723,"words":724,"published":725,"date":156},"\u002Fposts\u002F2026\u002F2026-01-03-论文实训草稿",10445,"2026-01-03 14:10:49",{"path":727,"words":728,"published":729,"date":156},"\u002Fposts\u002F2026\u002F2026-01-06-人工智能导论",11461,"2026-01-07 03:25:49",{"path":731,"words":732,"published":733,"date":156},"\u002Fposts\u002F2026\u002F2026-01-07-机器学习相关算法",82,"2026-01-08 02:28:13",{"path":735,"words":736,"published":737,"date":156},"\u002Fposts\u002F2026\u002F2026-01-16-关于低复杂度最长链共识算法设计",4466,"2026-01-16 10:33:23",{"path":739,"words":740,"published":741,"date":156},"\u002Fposts\u002F2026\u002F2026-01-19-杂记2026-01-19",5754,"2026-01-20 01:03:26",{"path":743,"words":744,"published":745,"date":156},"\u002Fposts\u002F2026\u002F2026-01-22-github-action-自动同步博客到-github-主页",1069,"2026-01-22 16:31:47",{"path":747,"words":748,"published":749,"date":156},"\u002Fposts\u002F2026\u002F2026-03-01-windows-wsl安装docker",257,"2026-03-01 20:55:31",{"path":751,"words":752,"published":753,"date":156},"\u002Fposts\u002F2026\u002F2026-03-14-关于-msys2",701,"2026-03-15 03:05:04",{"path":755,"words":756,"published":757,"date":156},"\u002Fposts\u002F2026\u002F2026-03-28-团队项目协作规范随记",2419,"2026-03-29 05:33:50",{"path":759,"words":760,"published":761,"date":156},"\u002Fposts\u002F2026\u002F2026-03-29-oh-my-posh配置分享",104,"2026-03-29 16:34:43",{"path":763,"words":764,"published":765,"date":156},"\u002Fposts\u002F2026\u002F2026-06-13-从上下文工程到-agent-harness-engineering",8840,"2026-06-13 21:06:20",{"path":767,"words":768,"published":769,"date":156},"\u002Fposts\u002F2026\u002F2026-07-13-简谈前端基建质量标准与ai友好建设",3814,"2026-07-13 21:06:20",{"path":771,"words":772,"published":773,"date":156},"\u002Fposts\u002F2026\u002F2026-07-15-agentic-rag-系统实践",13357,"2026-07-15 21:06:20",{"path":775,"words":776,"published":777,"date":156},"\u002Fposts\u002F2026\u002F2026-07-18-简谈sdd规范驱动开发-copy",4476,"2026-07-18 11:06:20",{"path":779,"words":780,"published":781,"date":156},"\u002Fposts\u002F2026\u002F2026-07-20-个人neovim配置分享",366,"2026-07-22 13:46:50",{"path":783,"words":784,"published":785,"date":156},"\u002Fposts\u002F2026\u002F2026-07-20-关于技术写作-设计文档",6763,"2026-07-20 13:46:50",{"path":787,"words":788,"published":789,"date":156},"\u002Fposts\u002F2026\u002F2026-07-29-fastapi-工程实践随记",3417,"2026-07-29 04:58:34",{"path":791,"words":792,"published":793,"date":156},"\u002Fposts\u002Frfc_agentic_project\u002F2026-06-19-rfc-liskin_code_agent",3563,"2026-06-19 15:18:20",{"path":795,"words":796,"published":797,"date":156},"\u002Fposts\u002Frfc_agentic_project\u002F2026-07-25-rfc-glushkov_ogas",13461,"2026-07-28 23:46:50",{"path":799,"words":800,"published":793,"date":156},"\u002Fposts\u002Frfc_agentic_project\u002F2026-08-03-adr-ogas-arkhiv",2698,{"path":802,"words":803,"published":804,"date":156},"\u002Fposts\u002Frfc_agentic_project\u002Frag\u002F2026-08-12-arkhiv_rag-vol-1",4366,"2026-08-12 15:18:20",{"path":806,"words":807,"published":808,"date":156},"\u002Fposts\u002Fleetcode\u002F2025-12-25-力扣百题速练javascripttypescriptvol-1",3108,"2025-12-25 13:57:08",{"path":810,"words":811,"published":812,"date":156},"\u002Fposts\u002Fleetcode\u002F2025-12-26-力扣百题速练javascripttypescriptvol-2",3185,"2025-12-26 13:57:08",{"path":814,"words":815,"published":816,"date":156},"\u002Fposts\u002Fleetcode\u002F2025-12-28-力扣百题速练javascripttypescriptvol-3",2533,"2025-12-29 02:15:44",{"path":818,"words":819,"published":820,"date":156},"\u002Fposts\u002Fleetcode\u002F2026-01-17-力扣百题速练javascripttypescriptvol-4",3024,"2026-01-17 22:53:01",{"path":822,"words":823,"published":824,"date":156},"\u002Fposts\u002Fleetcode\u002F2026-01-19-力扣百题速练javascripttypescriptvol-5",293,"2026-01-19 15:15:12",{"path":826,"words":827,"published":828,"date":156},"\u002Fposts\u002F分布式处理与计算\u002F分布式处理与计算spark系统与scala语言",4320,"2026-07-06 14:10:49",{"path":830,"words":831,"published":832,"date":156},"\u002Fposts\u002F分布式处理与计算\u002F分布式处理与计算分类与回归分析",2081,"2026-07-07 12:40:29",{"path":834,"words":835,"published":836,"date":156},"\u002Fposts\u002F分布式处理与计算\u002F分布式处理与计算数值优化方法",2182,"2026-07-07 13:50:29",{"path":838,"words":839,"published":840,"date":156},"\u002Fposts\u002F分布式处理与计算\u002F分布式处理与计算数据降维",1583,"2026-07-07 21:50:29",{"path":842,"words":843,"published":844,"date":156},"\u002Fposts\u002F分布式处理与计算\u002F分布式处理与计算聚类分析",1882,"2026-07-07 16:50:29",{"path":846,"words":847,"published":848,"date":156},"\u002Fposts\u002F分布式处理与计算\u002F分布式处理与计算随机模拟与统计推断",2433,"2026-07-06 22:40:29",{"path":850,"words":851,"published":852,"date":156},"\u002Fposts\u002F数据可视化\u002F2026-06-14-数据可视化-基本图像",10011,"2026-06-14 20:18:20",{"path":854,"words":855,"published":856,"date":156},"\u002Fposts\u002F数据可视化\u002F2026-06-14-数据可视化-概论",10030,"2026-06-14 15:18:20",{"path":858,"words":859,"published":860,"date":156},"\u002Fposts\u002F数据可视化\u002F2026-06-15-数据可视化-可视化库",2032,"2026-06-15 20:18:20",{"path":862,"words":863,"published":864,"date":156},"\u002Fposts\u002F数据结构\u002F2025-09-29-数据结构-图",5451,"2025-04-15 20:48:06",{"path":866,"words":867,"published":868,"date":156},"\u002Fposts\u002F数据结构\u002F2025-09-29-数据结构-树",4211,"2025-04-30 02:19:06",{"path":870,"words":871,"published":872,"date":156},"\u002Fposts\u002F数据结构\u002F2025-09-29-数据结构-队列",921,"2025-04-11 15:30:08",{"path":874,"words":875,"published":876,"date":156},"\u002Fposts\u002F数据结构\u002F关于数据结构的一些想法",456,"2025-04-04 19:13:49",{"path":878,"words":879,"published":880,"date":156},"\u002Fposts\u002F数据结构\u002F数据结构-栈",920,"2025-04-18 00:17:37",{"path":882,"words":883,"published":884,"date":156},"\u002Fposts\u002F数据结构\u002F数据结构-算法复杂度",2153,"2025-04-05 16:09:55",{"path":886,"words":887,"published":888,"date":156},"\u002Fposts\u002F数据结构\u002F数据结构-线性表链表",4572,"2025-04-06 00:24:40",{"path":890,"words":891,"published":892,"date":156},"\u002Fposts\u002F数据结构\u002F数据结构-线性表顺序表",4191,"2025-04-05 22:09:59",{"path":894,"words":895,"published":896,"date":156},"\u002Fposts\u002F数据结构\u002F数据结构-绪论",2881,"2025-04-05 11:31:22",{"path":898,"words":899,"published":156,"date":900},"\u002Fposts\u002F杂记\u002F2023-12-26-2023年12月26日",7,"2023-12-26 23:51:33",{"path":902,"words":899,"published":156,"date":903},"\u002Fposts\u002F杂记\u002F2024-01-01-2024团建","2024-01-01 01:51:27",{"path":905,"words":581,"published":156,"date":906},"\u002Fposts\u002F杂记\u002F2024-01-09-随记","2024-01-09 20:50:35",{"path":908,"words":909,"published":156,"date":910},"\u002Fposts\u002F杂记\u002F2024-03-12-花名",5275,"2024-03-12 16:04:23",{"path":912,"words":913,"published":156,"date":914},"\u002Fposts\u002F杂记\u002F2024-04-15-逆模因",1353,"2024-04-15 00:26:16",{"path":916,"words":917,"published":156,"date":918},"\u002Fposts\u002F杂记\u002F2024-05-03-艺作",27,"2024-05-03 00:01:00",{"path":920,"words":921,"published":156,"date":922},"\u002Fposts\u002F杂记\u002F2024-05-21-关于博客",1720,"2024-02-21 23:21:11",{"path":924,"words":925,"published":156,"date":926},"\u002Fposts\u002F杂记\u002F2024-06-20-dnvrc维克多崔专访-2023",2036,"2024-06-20 00:28:39",{"path":928,"words":760,"published":156,"date":929},"\u002Fposts\u002F杂记\u002F2024-06-25-6月25日随记","2024-06-25 12:46:36",{"path":931,"words":932,"published":156,"date":933},"\u002Fposts\u002F杂记\u002F2024-06-25-歌名-想見星星",148,"2024-06-25 22:47:59",{"path":935,"words":936,"published":156,"date":937},"\u002Fposts\u002F杂记\u002F2024-07-11-桂花岗",6,"2024-07-11 00:28:17",{"path":939,"words":940,"published":156,"date":941},"\u002Fposts\u002F杂记\u002F2024-07-15-rolling-bocci",9,"2024-07-15 21:25:56",{"path":943,"words":944,"published":156,"date":945},"\u002Fposts\u002F杂记\u002F2024-07-22-七月九日",34,"2024-07-09 17:54:23",{"path":947,"words":948,"published":156,"date":949},"\u002Fposts\u002F杂记\u002F2024-07-22-六月随记",1052,"2024-07-22 01:03:19",{"path":951,"words":952,"published":156,"date":953},"\u002Fposts\u002F杂记\u002F2024-07-22-石菖蒲",41,"2024-07-10 01:26:05",{"path":955,"words":514,"published":156,"date":956},"\u002Fposts\u002F杂记\u002F2024-07-23-新书","2024-07-23 01:05:49",{"path":958,"words":959,"published":156,"date":960},"\u002Fposts\u002F杂记\u002F2024-07-29-年前",74,"2024-07-29 23:59:07",{"path":962,"words":963,"published":156,"date":964},"\u002Fposts\u002F杂记\u002F2024-07-30-周边",138,"2024-07-30 02:03:23",{"path":966,"words":967,"published":156,"date":968},"\u002Fposts\u002F杂记\u002F2024-07-31-2024-7-31日",185,"2024-07-31 01:21:52",{"path":970,"words":971,"published":156,"date":972},"\u002Fposts\u002F杂记\u002F2024-08-02-八月",1710,"2024-08-30 00:31:02",{"path":974,"words":975,"published":156,"date":976},"\u002Fposts\u002F杂记\u002F2024-09-01-九月",1484,"2024-09-26 16:30:15",{"path":978,"words":979,"published":156,"date":980},"\u002Fposts\u002F杂记\u002F2024-09-08-2024年9月8日",97,"2024-09-08 13:59:36",{"path":982,"words":514,"published":156,"date":983},"\u002Fposts\u002F杂记\u002F2024-09-19-博客站的一周年","2024-09-18 23:36:24",{"path":985,"words":986,"published":156,"date":987},"\u002Fposts\u002F杂记\u002F2024-10-03-十月",1028,"2024-10-31 23:54:27",{"path":989,"words":990,"published":156,"date":991},"\u002Fposts\u002F杂记\u002F2024-11-04-十一月",2134,"2024-11-30 00:04:39",{"path":993,"words":994,"published":156,"date":995},"\u002Fposts\u002F杂记\u002F2024-11-10-gzhu-site-193",80,"2024-11-10 19:19:12",{"path":997,"words":998,"published":156,"date":999},"\u002Fposts\u002F杂记\u002F2024-12-01-十二月",455,"2024-12-30 23:13:47",{"path":1001,"words":1002,"published":156,"date":1003},"\u002Fposts\u002F杂记\u002F2024-12-21-年末",23,"2024-12-21 23:21:49",{"path":1005,"words":522,"published":156,"date":1006},"\u002Fposts\u002F杂记\u002F2025-01-30-一月","2025-01-30 23:44:41",{"path":1008,"words":1009,"published":156,"date":1010},"\u002Fposts\u002F杂记\u002F2025-03-01-二月",224,"2025-02-28 00:11:08",{"path":1012,"words":1013,"published":156,"date":1014},"\u002Fposts\u002F杂记\u002F2025-03-01-博客杂谈",344,"2025-03-01 16:03:18",{"path":1016,"words":1017,"published":156,"date":1018},"\u002Fposts\u002F杂记\u002F2025-03-05-三月",2090,"2025-03-05 19:17:28",{"path":1020,"words":1021,"published":156,"date":1022},"\u002Fposts\u002F杂记\u002F2025-07-10-七月",361,"2025-07-10 22:04:32",{"path":1024,"words":1025,"published":156,"date":1026},"\u002Fposts\u002F杂记\u002F2025-09-21-九月",781,"2025-09-21 20:53:37",{"path":1028,"words":1029,"published":156,"date":1030},"\u002Fposts\u002F杂记\u002F2025-10-09-十月",591,"2025-10-09 00:35:07",{"path":1032,"words":1033,"published":156,"date":1034},"\u002Fposts\u002F杂记\u002Fcontrol",13,"2023-09-25 13:03:59",{"path":1036,"words":1037,"published":156,"date":1038},"\u002Fposts\u002F杂记\u002Fquadim四叉树图像处理",1094,"2023-10-21 11:27:06",{"path":1040,"words":1041,"published":156,"date":1042},"\u002Fposts\u002F杂记\u002Fsuccess",24,"2023-09-22 15:59:07",{"path":1044,"words":1045,"published":156,"date":1046},"\u002Fposts\u002F杂记\u002F三光年",295,"2023-11-18 04:18:40",{"path":1048,"words":1049,"published":156,"date":1050},"\u002Fposts\u002F杂记\u002F加密测试",961,"2023-11-21 14:24:01",{"path":1052,"words":1053,"published":156,"date":1054},"\u002Fposts\u002F杂记\u002F啊啊啊啊啊啊啊",54,"2023-09-20 17:29:33",{"path":1056,"words":1057,"published":156,"date":1058},"\u002Fposts\u002F杂记\u002F杂谈",124,"2023-09-25 23:11:32",{"path":1060,"words":1061,"published":156,"date":1062},"\u002Fposts\u002F杂记\u002F艹",25,"2023-10-21 17:40:43",{"path":1064,"words":1065,"published":1066,"date":156},"\u002Fposts\u002F神经网络\u002F神经网络与深度学习卷积神经网络",3168,"2026-06-07 11:23:08",{"path":1068,"words":1069,"published":1070,"date":156},"\u002Fposts\u002F神经网络\u002F神经网络与深度学习循环神经网络",2348,"2026-06-07 11:33:08",{"path":1072,"words":1073,"published":1074,"date":156},"\u002Fposts\u002F神经网络\u002F神经网络与深度学习改进学习方法",3360,"2026-06-07 10:13:08",{"path":1076,"words":1077,"published":1078,"date":156},"\u002Fposts\u002F神经网络\u002F神经网络与深度学习深度生成模型",2727,"2026-06-07 11:43:08",{"path":1080,"words":1081,"published":1082,"date":156},"\u002Fposts\u002F神经网络\u002F神经网络与深度学习神经网络基础",3205,"2026-06-07 10:03:08",{"id":1084,"title":1085,"abbrlink":1086,"body":1087,"category":5,"cover":156,"date":156,"description":1545,"extension":1546,"mathjax":1547,"meta":1548,"navigation":1547,"path":727,"published":729,"readingMinutes":1551,"seo":1552,"stem":1553,"sticky":156,"swiper_index":156,"tags":1554,"updated":156,"words":728,"__hash__":1555},"posts\u002Fposts\u002F2026\u002F2026-01-06-人工智能导论.md","2026-01-06-人工智能导论笔记","64338",{"type":1088,"value":1089,"toc":1541},"minimark",[1090,1094,1098,1101,1104,1107,1111,1114,1117,1120,1123,1126,1134,1137,1140,1143,1147,1150,1153,1156,1159,1170,1208,1212,1215,1249,1337,1340,1344,1351,1354,1360,1387,1391,1394,1424,1427,1430,1493,1519,1523,1526,1529,1532,1535,1538],[1091,1092,1093],"h1",{"id":1093},"人工智能的基本概念与研究范畴",[1095,1096,1097],"p",{},"要理解人工智能，首先要从人类智能谈起。",[1095,1099,1100],{},"人类智能是人在认识、适应和改造客观世界的过程中，由一系列核心能力构成的综合性心智功能，其本质在于能够理解、推理、学习并运用知识去解决复杂问题。这种智能表现出若干鲜明的特点：人能够通过感官获取外部信息并理解其含义，能够从经验中不断学习、更新知识以适应新的环境，能够运用逻辑、归纳与演绎等方法从已知推知未知并制定解决问题的策略；与此同时，人还借助语言这一复杂的符号系统进行交流、表达和思考，并且具有自我意识与能动性，能够开展有目的、有计划的主动行为。正是这些能力的综合，构成了我们所说的\"智能\"。",[1095,1102,1103],{},"人工智能正是以模拟人类的上述能力为出发点的一门学科。它是计算机科学的一个分支，旨在研究并开发用于模拟、延伸和扩展人的智能的理论、方法、技术及应用系统，其根本目标是让机器能够胜任那些通常需要人类智能才能完成的复杂工作。人工智能的发展并非一蹴而就，而是经历了几个相互衔接的阶段。在 1956 年以前的孕育期，数理逻辑、控制论、信息论等理论为这门学科的诞生奠定了基础，图灵更是提出了\"机器能思考吗？\"这一划时代的问题。1956 年的达特茅斯会议正式确立了\"人工智能\"这一学科名称，此后进入形成与热潮期，研究者在问题求解、定理证明、机器翻译等方向取得了突破，乐观情绪一度高涨。进入二十世纪七八十年代，人们逐渐意识到\"知识\"的关键作用，专家系统将人类专家的知识规则化，成为这一时期的主流，人工智能也由此走向商业化应用。此后随着互联网的兴起和数据量的增长，以统计学习和神经网络复兴为代表的机器学习方法逐渐成为核心，支持向量机、决策树等方法得到广泛应用。而进入 2010 年代以后，得益于大数据、以 GPU 为代表的强大算力以及深度神经网络等算法改进，深度学习在图像识别、自然语言处理等领域取得突破性进展，从而引发了新一轮的人工智能热潮。",[1095,1105,1106],{},"人工智能大致涵盖几个彼此关联的方面。知识表示研究如何用机器可处理的形式来表示和存储人类的事实与规则；机器感知研究如何让机器通过\"感官\"获取外部信息，其核心是承担\"看\"的计算机视觉与承担\"听\"的语音识别；在感知的基础上，机器思维进一步研究如何对信息进行推理、决策与问题求解，这构成了人工智能的核心；机器学习则研究如何让机器自动从数据中学习规律和知识，从而不断改进自身性能，是实现智能的关键途径；自然语言处理关注人机之间自然语言通信的理解与生成；最终，这些能力被综合起来，用以构建能够对外界环境做出合理反应与行动的智能体或机器人。围绕这些内容，人工智能形成了若干活跃的研究领域，包括以计算机视觉、语音识别和多模态感知为代表的机器感知，以机器翻译、文本理解与生成、对话系统为代表的自然语言处理，以深度学习、强化学习、迁移学习、联邦学习为代表的机器学习，以知识图谱、自动推理、专家系统为代表的知识表示与推理，以及涉及环境感知、运动控制、人机协作的机器人学；此外，人工智能还在智能驾驶、演化计算与群智能优化、数据挖掘与大数据分析，以及生物信息学、计算金融、智慧医疗等交叉领域不断拓展。",[1091,1108,1110],{"id":1109},"第二章-知识表示","第二章 知识表示",[1095,1112,1113],{},"如果说人工智能以运用知识为核心，那么首先要厘清什么是知识。",[1095,1115,1116],{},"知识是经过加工、整理、解释、挑选和改造的信息，是人们在长期实践中积累起来的对客观世界的规律性认识，在人工智能中它是使机器具备智能的基石。知识具有若干重要特性：它具有相对正确性，即往往只在特定条件下成立；它带有不确定性；它是可表示的，也是可利用的。人们可以从不同角度对知识进行分类，按作用层次可分为事实性知识、过程性知识与控制性知识，按确定性可分为确定性知识与不确定性知识，按表现形式则可分为可编码的显性知识与依赖经验、直觉的隐性知识。",[1095,1118,1119],{},"将这些知识加以形式化、模型化，使计算机能够存储、处理和运用，就是知识表示所要解决的问题，它本质上是数据结构与解释过程的结合。选择何种知识表示方法并非随意，而要综合考虑几个方面：能否充分表示领域知识、能否支持高效推理、是否便于知识的获取与管理，以及是否易于理解和维护。",[1095,1121,1122],{},"在众多表示方法中，逻辑表示法具有基础性的地位。命题是一个能够判断其真或假的陈述句，例如\"太阳从东方升起\"为真，而\"1 加 1 等于 3\"为假。命题逻辑虽然简洁，却无法刻画命题内部的结构，于是谓词应运而生。谓词用于描述个体的性质或个体之间的关系，例如\"是红色的(x)\"描述性质，\"朋友(x, y)\"描述关系。其中，能够独立存在的具体或抽象对象称为个体，个体所组成的集合即讨论范围称为个体域。这里需要特别区分函数与谓词：函数的返回值是一个个体，如\"父亲(小明)\"返回一个人；而谓词的返回值是一个真值，如\"朋友(小明, 小红)\"用以判断真假。命题逻辑是谓词逻辑的基础，谓词逻辑则是命题逻辑的细化和扩展，二者都使用逻辑连接词和真值运算，但谓词逻辑能够深入到命题内部，通过个体、谓词和量词刻画更精细的结构。",[1095,1124,1125],{},"在谓词逻辑中，充当逻辑自变量的个体称为项，项可以是常量、变量或函数；谓词的阶由项的取值范围决定，一阶谓词的项是个体，二阶谓词的项则可以是谓词或集合；",[1095,1127,1128,1129,1133],{},"谓词的一般形式可写作 ",[1130,1131,1132],"code",{},"P(x1, x2, ..., xn)","，其中 P 是谓词名，各 x 为项。由谓词、项、逻辑连接词、量词和括号按规则组成的合法符号串称为谓词公式，而对公式中的个体常量、函数符号和谓词符号赋予具体含义、指定个体域与对应关系，就构成了公式的解释，由此确定公式的真值。一阶谓词逻辑表示法属于结构化知识表示，它精确地描述了知识内部的逻辑结构，尤其擅长表达事物的状态、属性、概念以及它们之间精确的逻辑关系，能够严格地刻画\"所有 A 都是 B\"\"存在某个 A 具有性质 P\"这类事实和规则。运用这种方法表示知识通常遵循这样的步骤：先定义个体域，确定讨论对象的集合；再定义谓词和函数，用符号表示个体的性质与关系；然后用连接词和量词将原子谓词公式组合成复合公式；最后对公式进行化简，化为前束范式一类的标准式，以便于后续推理。",[1095,1135,1136],{},"产生式表示法是另一类广泛使用的方法。产生式的基本形式是\"IF (前提) THEN (结论\u002F动作)\"，也称条件—行动对。它与谓词逻辑中的蕴涵式既有联系又有区别：产生式规则的前提是与动态数据库匹配的，匹配即执行，而蕴涵式表达的是静态的逻辑关系；产生式的 THEN 部分不仅可以断言新事实，还可以执行修改数据库、输出等动作，蕴涵式则仅表示逻辑推导；此外，产生式系统有独立的推理机来控制规则触发的顺序，逻辑系统则依赖通用的推理规则。一个完整的产生式系统由三部分组成，即存储全部规则的规则库、存储当前事实与中间结论的综合数据库，以及负责匹配、冲突消解和执行的推理机。它求解问题的一般过程是：先将初始事实存入综合数据库，推理机随后把规则的前提与数据库中的事实逐一比对，若有多条规则同时满足则按某种策略进行冲突消解，接着执行被选中规则的结论部分以更新数据库，如此循环往复，直到达到目标状态或再无规则可用为止。推理机的推理方式包括从已知事实出发的正向推理、从假设目标出发的反向推理，以及二者结合的混合推理；当发生冲突时，可以采用专一性排序、规则排序、数据排序、就近排序、规模排序等策略来选择应当激活的规则。总体而言，产生式表示法结构清晰、模块化程度高、便于表达因果与经验性知识，但在处理结构化关系和大规模规则时也存在效率和一致性方面的局限。",[1095,1138,1139],{},"框架表示法则从另一个角度组织知识，它是一种描述固定、典型情景中对象的结构化表示。框架的一般形式以框架名为标识，其下由若干\"槽\"描述对象的属性或方面，每个槽又可配以若干\"侧面\"来说明属性的更详细信息，如默认值、取值范围或触发过程，并可附加约束条件。框架表示法的突出特点在于结构性好、支持通过 AKO 槽实现的继承性、符合人们对典型事物认知的自然性，以及便于表达默认知识。运用框架表示知识时，通常先分析待描述对象并确定框架名，再确定关键属性作为槽，为每个槽配备相应的侧面，填入具体的值，并最终确定它与其它框架之间通过 AKO、ISA 等槽体现的继承关系。譬如描述一间卧室，可以以\"卧室\"为框架名，通过 AKO 指明它是\"房间\"的一种，再用位置、面积、功能、包含家具等槽刻画其属性，并进一步以\"床\"\"书桌\"等子框架描述家具的类型、材质、状态与摆放物品；同样，描述一台计算机主机时，也可以\"计算机主机\"为框架名，AKO 指向\"电子设备\"，用品牌、型号、状态等槽记录整机信息，再以处理器、内存、硬盘、主板、电源等槽及其侧面刻画各个组件的具体参数。",[1095,1141,1142],{},"近年来兴起的知识图谱可以看作是一种以图结构表示实体及其关系的语义网络。以一个关于《红楼梦》的微型知识图谱为例，\"曹雪芹\"\"《红楼梦》\"\"贾宝玉\"\"林黛玉\"\"大观园\"等构成图中的节点，也就是实体，而\"创作\"\"文学体裁\"\"主角\"\"居住地\"\"爱慕\"等则构成连接节点的边，也就是关系。以三元组形式加以组织，便能清晰地表达出实体之间丰富而结构化的语义联系。",[1091,1144,1146],{"id":1145},"第三章-确定性推理","第三章 确定性推理",[1095,1148,1149],{},"推理是从已知事实出发、运用知识推出结论的思维过程，是人工智能实现\"思维\"的核心环节。按照信息流动的方向，推理可以分为几种不同的方式。正向推理由数据驱动，从已知事实出发匹配规则，不断推出新事实直至到达目标，它适用于初始数据明确、目标众多的场合，但过程中可能进行大量与目标无关的推理。逆向推理则由目标驱动，从假设的目标出发反向寻找支持它的证据，其目的性强、适用于目标单一的情形，但对初始数据的指导性较弱。混合推理综合了二者的长处，先从初始事实正向推理得到中间结论，再从目标出发逆向寻求支持，因而效率更高。",[1095,1151,1152],{},"在产生式系统的推理过程中，常常会遇到冲突，即在某一时刻有多条规则的前提同时与综合数据库匹配成功。解决冲突需要一定的策略：可以采用专一性排序，优先使用条件更具体、范围更小的规则；可以采用规则排序，按事先固定的优先级选择；也可以采用数据排序，根据匹配事实的新旧程度或特定性来判断；还可以采用就近排序，优先使用最近被触发过的规则。",[1095,1154,1155],{},"在自动定理证明中，归结推理占有重要地位，而它以子句和子句集为基本对象。子句是若干文字的析取式，其中每个文字或为原子公式或为其否定；子句集则是若干子句的集合，本质上是一个合取范式，即子句之间以\"与\"相连。",[1095,1157,1158],{},"将任意谓词公式化为子句集需要经过一系列变换：",[1095,1160,1161,1162,1165,1166,1169],{},"先消去蕴涵符号，用 ",[1130,1163,1164],{},"¬A ∨ B"," 替换 ",[1130,1167,1168],{},"A → B","；再将否定符内移到原子公式之前；接着对变量进行标准化，使不同量词约束的变量互不同名；然后通过 Skolem 化消去存在量词；再化为前束形，把所有全称量词移到公式最前面；随后化为合取范式；最后消去全称量词和合取词，得到子句集。需要注意的是，谓词公式与它的子句集并不完全等价，因为在转化过程中，尤其是 Skolem 化会引入新的常量或函数，二者只是在不可满足性上等价——原公式不可满足当且仅当其子句集不可满足，这一点恰恰是归结原理得以用于自动定理证明的理论基础。",[1095,1171,1172,1173,1176,1177,1180,1181,1184,1185,1188,1189,1192,1193,1196,1197,1199,1200,1203,1204,1207],{},"鲁宾孙归结原理的引入，为定理的机器自动证明提供了一种简洁、规范且完备的推理方法，它把复杂的推理归结为简单的子句归结，从而奠定了自动推理的理论基础。所谓归结原理，是在子句集中进行操作：若两个子句分别包含一对互补文字，如 ",[1130,1174,1175],{},"P"," 与 ",[1130,1178,1179],{},"¬P","，便可消去这对互补文字，将两子句的其余部分合并构成新的子句，这一新子句即称为归结式。利用归结原理求解问题的一般步骤是：先把已知条件表示为谓词公式集合 ",[1130,1182,1183],{},"F","，把待证目标表示为公式 ",[1130,1186,1187],{},"G","，再将 ",[1130,1190,1191],{},"{F, ¬G}"," 化为子句集 ",[1130,1194,1195],{},"S","，随后对 ",[1130,1198,1195],{}," 反复应用归结原理；若最终能推出空子句 ",[1130,1201,1202],{},"□","，就说明 ",[1130,1205,1206],{},"F → G"," 成立，证明完成；反之，若无法归结出空子句且无法继续归结，则说明原结论不成立。",[1091,1209,1211],{"id":1210},"第四章-不确定性推理","第四章 不确定性推理",[1095,1213,1214],{},"现实世界中的知识与证据往往并不精确、完备，甚至彼此矛盾，因而单靠确定性推理难以应对。不确定性推理正是在知识和证据不精确、不完备、模糊或存在矛盾的情况下，依然能够进行推理并得出结论的方法，其结论通常附带有不确定性的度量。围绕这一目标，人们发展出可信度方法、主观贝叶斯方法、证据理论以及模糊推理等多种途径。无论采用哪种方法，都需要解决几个基本问题：如何表示知识、证据和结论的不确定性度量，如何定义不确定性在推理过程中的传播、更新与组合，如何明确这些度量的实际语义，以及当多条证据支持同一结论时如何合理地加以组合。",[1095,1216,1217,1218,1221,1222,1225,1226,1229,1230,1233,1234,1236,1237,1240,1241,1244,1245,1248],{},"可信度方法是较早也较为直观的一种。可信度用来表示在证据成立的前提下假设为真的信任程度，其核心是可信度因子 ",[1130,1219,1220],{},"CF(H,E)","，它被定义为信任的增长度，即 ",[1130,1223,1224],{},"CF(H,E) = MB(H,E) - MD(H,E)","，其中 ",[1130,1227,1228],{},"MB"," 表示证据对假设的信任增加度量，",[1130,1231,1232],{},"MD"," 表示不信任增加度量。当 ",[1130,1235,1220],{}," 大于零时，说明证据的出现增加了假设为真的信任度，其值越大信任增加越多；等于零则表示证据与假设无关，或信任与不信任的增量相互抵消；小于零则表示证据反而增加了对假设为假的不信任。在实际求取结论可信度时，一般先以 ",[1130,1238,1239],{},"IF...THEN..."," 的形式建立规则库并为每条规则赋予可信度因子，再为初始证据赋予初始可信度，然后从已知证据出发进行正向推理；对于形如 ",[1130,1242,1243],{},"IF E THEN H"," 且带有规则可信度的规则，结论的可信度由前提的可信度与规则的可信度共同决定，通常取 ",[1130,1246,1247],{},"CF(H) = CF(E) * CF(Rule)","；如果同一结论由多条路径推出，还需借助合成公式将多个可信度组合为最终结果。",[1095,1250,1251,1252,1255,1256,1259,1260,1262,1263,1266,1267,1270,1271,1274,1275,1277,1278,1281,1282,1285,1286,1288,1289,1292,1293,1295,1296,1299,1300,1303,1304,1306,1307,1310,1311,1313,1314,1316,1317,1320,1321,1324,1325,1328,1329,1332,1333,1336],{},"证据理论，即 D-S 理论，提供了另一套刻画不确定性的框架，其中有三个核心概念。设识别框架 ",[1130,1253,1254],{},"Θ"," 为所有可能假设的集合。概率分配函数 ",[1130,1257,1258],{},"m"," 是从 ",[1130,1261,1254],{}," 的幂集到区间 ",[1130,1264,1265],{},"[0,1]"," 的映射，满足 ",[1130,1268,1269],{},"m(∅)=0"," 且各 ",[1130,1272,1273],{},"m(A)"," 之和为 1，",[1130,1276,1273],{}," 表示证据本身对命题 ",[1130,1279,1280],{},"A"," 的基本概率分配。信任函数 ",[1130,1283,1284],{},"Bel"," 对任意命题 ",[1130,1287,1280],{}," 定义为其所有子集基本概率之和，即 ",[1130,1290,1291],{},"Bel(A) = ∑_{B ⊆ A} m(B)","，表示证据对 ",[1130,1294,1280],{}," 的总信任度。似然函数 ",[1130,1297,1298],{},"Pl"," 则定义为 ",[1130,1301,1302],{},"Pl(A) = 1 - Bel(¬A)","，表示不否定 ",[1130,1305,1280],{}," 的程度，区间 ",[1130,1308,1309],{},"[Bel(A), Pl(A)]"," 构成信任区间，刻画了对 ",[1130,1312,1280],{}," 的不确定程度。需要强调的是，概率分配函数与经典概率并不相同：其一，概率分配函数的定义域是识别框架的幂集，即所有子集，而经典概率的定义域是基本事件本身；其二，",[1130,1315,1273],{}," 可以直接分配给任意命题，无需再将支持度细分到内部元素，而经典概率必须满足可加性；其三，证据理论允许 ",[1130,1318,1319],{},"m(Θ) > 0","，即保留一部分信任度用以表示\"未知\"，而经典概率中 ",[1130,1322,1323],{},"P(Θ)=1"," 是确定的，无法表达这种无知。基于此，D-S 理论用基本概率分配函数描述证据，用信任函数和似然函数构成的信任区间描述假设，规则则可表示为带有概率分配的条件形式；当有两个相互独立的证据源时，通过 Dempster 组合规则实现推理组合，其组合后的概率分配为 ",[1130,1326,1327],{},"m(C) = K^{-1} * ∑_{A∩B=C} m1(A)*m2(B)","（",[1130,1330,1331],{},"C ≠ ∅","），其中归一化常数 ",[1130,1334,1335],{},"K = 1 - ∑_{A∩B=∅} m1(A)*m2(B)"," 用于排除冲突证据的影响。",[1095,1338,1339],{},"除随机性引起的不确定性之外，还有一类源于概念本身的不确定性，这就是模糊性。模糊性指事物在概念和外延上所具有的不分明性，源于事物类属的\"亦此亦彼\"，是对静态事物本身状态的描述。它与随机性有着本质区别：随机性源于因果律的缺失，刻画的是事件是否发生的不确定性，用概率论来描述；而模糊性源于排中律的缺失，刻画的是事件本身状态的不确定性，用模糊集合论来描述。譬如\"明天可能下雨\"体现的是随机性，而\"现在是阴天\"体现的是模糊性，因为\"阴天\"本身的边界并不分明。日常生活中诸如高个子、年轻人、热水、天气很好、有点咸、打扫干净之类的说法，都是典型的模糊概念。处理这类问题的模糊推理一般经历五个环节：首先是模糊化，依据预先定义的隶属度函数把精确的输入值转化为对应模糊语言变量的隶属度；其次是模糊规则匹配，将模糊化后的输入与知识库中的模糊规则前件比对，用取小或乘积运算计算每条规则的激活强度；然后是模糊推理，依据激活强度对规则后件对应的模糊集进行裁剪或缩放，得到各条规则的模糊结论；接着进行模糊结论合成，用取大运算把所有被激活规则的结论聚合为一个综合的输出模糊集合；最后是去模糊化，通过重心法、最大隶属度法或中位数法等算法把输出模糊集合转化为一个精确的输出值。",[1091,1341,1343],{"id":1342},"第五章-搜索求解","第五章 搜索求解",[1095,1345,1346,1347,1350],{},"许多人工智能问题最终都可以归结为在状态空间中寻找从初始状态到目标状态路径的过程，这一过程即为搜索。搜索方法大体可分为两类：一类是盲目搜索，它不利用额外信息，只按固定顺序展开；另一类是启发式搜索，它借助启发信息来指导搜索方向，二者的根本区别就在于是否使用启发信息。所谓启发信息，是关于问题领域的额外知识，常表示为启发函数 ",[1130,1348,1349],{},"h(n)","，用于估计从当前节点到目标的代价；启发式搜索正是利用这类信息来引导搜索、提高效率的方法。",[1095,1352,1353],{},"用状态空间法表示问题时，问题的解就是一个从初始状态到目标状态的操作序列，求解的本质即在状态空间中寻找路径，而其中总代价最小者称为最优解，值得注意的是最优解并不一定唯一。状态空间图的一般搜索过程可以描述为：先把初始状态放入 OPEN 表，若 OPEN 表为空则搜索失败，否则从中取出一个状态；若该状态即目标则搜索成功，否则将其扩展生成后继状态，并对后继状态加以处理，按既定策略把新状态放入 OPEN 表、将已扩展的父状态移入 CLOSED 表，如此反复直至成功或失败。在这一过程中，OPEN 表用于存放待考察的前沿节点，CLOSED 表用于存放已考察的历史节点，二者的区别正在于节点是待扩展还是已扩展。",[1095,1355,1356,1357,1359],{},"盲目搜索包含多种具体策略，如宽度优先搜索、深度优先搜索、一致代价搜索、深度受限搜索和迭代加深搜索。深度优先搜索本身没有所谓最优的子节点生成次序，通常按字母序等预设顺序展开；若希望引入启发性，可以按启发函数值排序，优先扩展 ",[1130,1358,1349],{}," 更小、即更接近目标的子节点。宽度优先搜索与深度优先搜索是两种典型而互补的策略：前者借助队列逐层扩展，后者借助栈沿分支深入；宽度优先搜索是完备的，并且在各步代价相等时能保证最优，但空间开销较大，深度优先搜索空间开销小，却可能不完备且找到的解未必最优。因此，当解位于较浅层、需要最优解且空间充足时，宽度优先搜索更为合适；而当搜索深度很大、只求可行解且宽度优先搜索的空间开销无法承受时，深度优先搜索则更具优势。",[1095,1361,1362,1363,1366,1367,1370,1371,1373,1374,1376,1377,1379,1380,1225,1383,1386],{},"在启发式搜索中，A* 搜索算法尤为重要。它以估价函数 ",[1130,1364,1365],{},"f(n) = g(n) + h(n)"," 来选择待扩展的节点，其中 ",[1130,1368,1369],{},"g(n)"," 为从初始节点到当前节点的实际代价，",[1130,1372,1349],{}," 为从当前节点到目标的估计代价，后者需要结合问题领域的知识来设计，例如以直线距离作为估计。A* 算法与一般 A 算法的区别在于：A 算法泛指一切使用 ",[1130,1375,1365],{}," 形式的算法，而 A* 算法是其特例，它额外要求 ",[1130,1378,1349],{}," 满足可采纳性，即 ",[1130,1381,1382],{},"h(n) ≤ h*(n)",[1130,1384,1385],{},"h*(n)"," 为真实的最小代价；正是在这一条件下，A* 算法才能保证找到最优解。",[1091,1388,1390],{"id":1389},"第六章-进化算法与群智能算法","第六章 进化算法与群智能算法",[1095,1392,1393],{},"除了基于逻辑与搜索的经典方法之外，受自然界启发的优化算法在求解复杂问题时同样占有重要位置，遗传算法便是其中的代表。遗传算法模拟生物进化的机制，其基本流程是：先随机生成由一组候选解构成的初始种群，计算每个个体的适应度值，再根据适应度高低选择优良个体作为父代；随后将父代两两配对，以一定概率交换部分基因产生子代，并以较低概率对子代的某些基因进行变异以引入新特征；接着用子代替换部分或全部父代形成新一代种群，如此循环，直至满足最大迭代次数或找到满意解等终止条件为止。遗传算法具有鲜明的特点：它对解空间中的多个点同时进行群体搜索，并行性好；它通过选择、交叉、变异等概率规则引导搜索，属于启发性的随机搜索；它只需目标函数的取值而无需梯度信息，不依赖函数的连续性与可微性；并且通过种群内个体的信息交换，隐含地实现了全局并行搜索。",[1095,1395,1396,1397,1400,1401,1404,1405,1408,1409,1411,1412,1415,1416,1419,1420,1423],{},"在遗传算法中，适应度函数扮演着核心角色。它一方面用来评价个体的优劣，适应度值的高低直接反映解的质量；另一方面又指导选择操作，适应度越高的个体被遗传到下一代的机会越大，从而成为算法进化的驱动力。构造适应度函数需结合具体问题：以在区间 ",[1130,1398,1399],{},"[0, 31]"," 上求 ",[1130,1402,1403],{},"f(x) = x²"," 的最大值为例，可用五位二进制串对 ",[1130,1406,1407],{},"x"," 编码，最直接的做法是将解码后的 ",[1130,1410,1407],{}," 代入目标函数，令 ",[1130,1413,1414],{},"fit(x) = x²"," 即为适应度；若要求最小值，则可构造 ",[1130,1417,1418],{},"fit(x) = C_max - f(x)"," 或 ",[1130,1421,1422],{},"fit(x) = 1 \u002F (f(x) + ε)","，以确保适应度非负且与目标值成反比。而选择操作的基本思想，正是模拟\"适者生存\"的自然法则，使适应度高的个体有更大概率被选中，将其优良基因传递给下一代，其核心是基于概率的优胜劣汰，引导搜索朝更优区域推进。",[1095,1425,1426],{},"在基本遗传算法的基础上，人们发展出若干改进形式。多种群遗传算法与基本遗传算法在基本操作单元和遗传算子上是一致的，但在种群结构上有所不同：基本遗传算法维护单一同质的种群，多种群遗传算法则并行维护多个子种群，并可定期在种群之间迁移优秀个体，其目的在于维持种群多样性、有效防止早熟收敛并增强全局探索能力，是并行遗传算法的一种典型实现。多倍体遗传算法同样遵循类似的进化流程，区别主要体现在基因编码上：基本遗传算法的个体是单倍体，只有一套基因，多倍体遗传算法的个体则具有多套基因；多倍体存在显隐性关系，个体的性状由显性基因决定，但遗传时两套基因共同参与，这种机制增强了算法的记忆能力以及对环境变化的鲁棒性。",[1095,1428,1429],{},"群智能算法则从另一条思路出发，它模拟鸟群、蚁群、鱼群等生物群体的集体智能行为。在这类算法中，简单的个体遵循相对简单的规则，个体之间以及个体与环境之间进行局部交互与信息共享，通过这些分散、自组织的局部互动，在群体层面涌现出复杂而高效的全局智能，从而求解复杂的优化或协同问题。群智能算法呈现出诸多共同特点：它们分布式、自组织，没有中心控制；它们既依靠正反馈来强化优质的解或路径，又依靠负反馈来避免陷入局部最优；由于个体简单，部分个体失效并不影响群体功能，因而鲁棒性强；个体可以同时独立行动，并行性好；同时它们对目标函数的要求很低，通用性好，适合黑箱优化。将群智能算法与进化算法相比较，二者都是受自然启发的元启发式优化方法，都属于群体智能范畴，但灵感来源不同——进化算法源于生物进化中的遗传、变异与选择，群智能算法则源于生物群体的协作、竞争与信息共享；核心操作也不同——进化算法以交叉、变异等基因操作为核心，群智能算法则以个体间的信息交互与行为模仿为核心；此外，进化算法的迭代代次界限分明，群智能算法中个体持续更新，代次的界限相对模糊。",[1095,1431,1432,1433,1436,1437,1440,1441,1444,1445,1448,1449,1451,1452,1454,1455,1458,1459,1462,1463,1466,1467,1470,1471,1474,1475,1478,1479,1482,1483,1485,1486,1488,1489,1492],{},"在群智能算法中，粒子群优化算法与蚁群优化算法最具代表性。粒子群算法可以用鸟群觅食来形象说明：一群鸟在某片区域随机搜寻食物，每只鸟都不知道食物确切的位置，但都记得自己以及同伴曾经找到的最佳位置，于是它们不断调整飞行方向和速度，既向自己记忆中的最优位置飞行，也向整个群体公认的最优位置靠拢，最终聚集到食物最丰富之处。其具体流程是：先随机初始化粒子的位置与速度并设定惯性权重、加速常数等参数，计算各粒子的适应度，再将每个粒子的当前位置与自身历史最优以及群体历史最优比较并更新，随后依据速度更新公式 ",[1130,1434,1435],{},"v_i(t+1) = w * v_i(t) + c1*r1*(pbest_i - x_i(t)) + c2*r2*(gbest - x_i(t))"," 和位置更新公式 ",[1130,1438,1439],{},"x_i(t+1) = x_i(t) + v_i(t+1)"," 更新粒子，直至达到精度要求或最大迭代次数为止。在速度更新方程中，各部分含义各不相同：惯性部分 ",[1130,1442,1443],{},"v_i(t)"," 代表对先前速度的继承，惯性权重 ",[1130,1446,1447],{},"w"," 用以平衡全局探索与局部开发，",[1130,1450,1447],{}," 大则探索能力强，",[1130,1453,1447],{}," 小则开发能力强；认知部分 ",[1130,1456,1457],{},"c1*r1*(pbest_i - x_i(t))"," 代表粒子向自身历史最优学习的趋势，由认知加速常数 ",[1130,1460,1461],{},"c1"," 控制个体经验的影响；社会部分 ",[1130,1464,1465],{},"c2*r2*(gbest - x_i(t))"," 代表粒子向群体历史最优学习的趋势，由社会加速常数 ",[1130,1468,1469],{},"c2"," 控制社会信息的影响；",[1130,1472,1473],{},"r1","、",[1130,1476,1477],{},"r2"," 则是增加随机性的随机数。粒子群算法原理简单、易于实现、参数少、收敛快，本质上是一种全局搜索，但后期容易陷入局部最优。它的寻优过程可分为前期广泛探索的探索阶段与后期精细搜索的开发阶段，其寻优准则包括全局最优在连续若干代内不再变化的收敛准则、达到预设最大迭代次数的迭代准则以及解的适应度达到目标的精度准则。在参数选择上，惯性权重常采用线性递减策略，初期取较大值（如 0.9）以利于探索、后期取较小值（如 0.4）以利于开发；加速常数通常取 ",[1130,1480,1481],{},"c1 = c2 = 2"," 左右，也有研究建议 ",[1130,1484,1461],{}," 由大到小、",[1130,1487,1469],{}," 由小到大变化；种群规模一般取 20 至 50，复杂问题可适当增大；速度上限 ",[1130,1490,1491],{},"V_max"," 通常设为变量范围的百分之十到二十，以防搜索步长过大。",[1095,1494,1495,1496,1499,1500,1503,1504,1507,1508,1511,1512,1515,1516,1518],{},"蚁群算法则可借助旅行商问题来说明其原理：蚂蚁最初随机选择路径并在沿途释放信息素，较短的路径由于蚂蚁往返更快、单位时间内信息素积累更多，后续蚂蚁便倾向于选择信息素浓度更高的路径，从而进一步强化这条路径；通过信息素的正反馈与挥发带来的负反馈相互配合，最终所有蚂蚁趋于收敛到最短路径之上。蚁群算法具有正反馈机制、能较快发现较好解，天然是分布式计算、易于并行，且启发性强、与具体问题结合紧密，但也存在初期信息素匮乏、收敛较慢以及参数设置对性能影响显著的不足。其寻优过程通常包含三个阶段：初始化阶段设定参数并初始化信息素；迭代构建阶段中每只蚂蚁根据路径上的信息素浓度和以距离倒数为代表的启发信息，以一定概率构建完整路径；信息素更新阶段在所有蚂蚁走完后，依据路径质量增加优质路径的信息素，同时让所有路径的信息素按比例挥发。它的寻优准则与粒子群算法类似，同样包括最大迭代次数、最优解连续不变的代数以及是否达到期望精度等。在参数选择方面，信息素重要性因子 ",[1130,1497,1498],{},"α"," 越大，蚂蚁越倾向于选择信息素浓的路径，收敛加快但易早熟；启发信息重要性因子 ",[1130,1501,1502],{},"β"," 越大，蚂蚁越倾向于选择看起来更近的路径，贪心性越强；信息素挥发系数 ",[1130,1505,1506],{},"ρ"," 取值于 ",[1130,1509,1510],{},"(0,1)"," 之间，值小则信息素留存久、全局搜索能力强但收敛慢，值大则挥发快、利于抛弃劣解但可能丢失历史信息；信息素强度 ",[1130,1513,1514],{},"Q"," 影响信息素增量的绝对值，与问题规模相关；蚂蚁数量 ",[1130,1517,1258],{}," 一般与问题节点数相当，过多会拖慢收敛，过少则正反馈不足。",[1091,1520,1522],{"id":1521},"第七章-人工智能的现状与未来","第七章 人工智能的现状与未来",[1095,1524,1525],{},"从逻辑推理、知识表示到搜索求解,再到进化与群智能算法,人工智能在其发展早期主要依赖人为设计的规则与显式编码的知识。这些方法奠定了学科的理论根基,却也在面对真实世界的复杂性、模糊性和海量数据时暴露出瓶颈。进入二十一世纪第二个十年以后,随着数据、算力和算法三者同时取得突破,人工智能进入了以数据驱动和深度学习为核心的新阶段,其面貌与经典时期相比已经发生了深刻的变化。",[1095,1527,1528],{},"从技术格局来看,当前人工智能的主流范式是深度神经网络。与依赖人工构造特征的传统方法不同,深度学习能够从原始数据中自动学习出多层次的特征表示,从而在图像识别、语音识别和自然语言处理等长期难以攻克的任务上取得了突破性进展。尤其是 2017 年提出的 Transformer 架构,凭借自注意力机制有效地建模了序列内部的长程依赖,成为此后大规模语言模型的技术基石。以此为基础,人们发现当模型参数规模、训练数据量和算力投入同步增长到一定程度时,模型会涌现出小模型所不具备的能力,这一现象被概括为\"规模定律\"。沿着这条路径,大规模预训练模型逐渐成为近年来最受关注的方向:模型先在海量无标注语料上进行自监督预训练以获得通用表征,再通过在下游任务上的微调或指令对齐适配到具体应用,由此形成了\"预训练加微调\"的通行范式。与此同时,人工智能的能力也从单一模态走向多模态,能够统一处理文本、图像、音频乃至视频,并在内容生成方面展现出前所未有的创造力,生成式人工智能因而成为当前产业和公众关注的焦点。",[1095,1530,1531],{},"技术的成熟带动了应用的广泛落地。智能推荐、机器翻译、语音助手和智能驾驶已经成为许多人习以为常的工具;在产业层面,人工智能被用于工业质检、药物发现、蛋白质结构预测、金融风控以及科学研究的辅助,并在其中一些领域达到甚至超越了人类专家的水平;在软件开发、内容创作、教育辅导等知识密集型工作中,以大模型为核心的智能助手也正在改变人们的工作方式。可以说,人工智能已经从实验室走向了社会经济的各个环节,由一项前沿技术逐渐演变为一种基础性的通用能力。",[1095,1533,1534],{},"繁荣之下仍然存在诸多值得清醒看待的局限与挑战。首先是可解释性问题,深度神经网络本质上是一个高度复杂的\"黑箱\",其决策过程难以被人类直观理解,这在医疗、司法、金融等高风险领域构成了严重障碍——这一点恰恰与前几章所讨论的、以逻辑和规则为基础、推理过程清晰可追溯的经典方法形成了鲜明对照。其次是数据依赖与成本问题,当前主流模型的性能高度依赖海量高质量数据和昂贵的算力,训练与部署的能耗和资金门槛都相当可观,这既带来了环境层面的担忧,也可能加剧技术资源的不平等。再次是可靠性问题,大模型存在\"幻觉\"现象,会以流畅自信的方式生成看似合理却与事实不符的内容,同时对输入扰动较为敏感、鲁棒性仍显不足。最后是伦理、安全与治理问题,数据隐私、算法偏见、虚假信息、版权争议以及潜在的滥用风险,都对社会治理提出了新的要求,如何在鼓励创新与防范风险之间取得平衡,已成为各国普遍面临的议题。",[1095,1536,1537],{},"面向未来,人工智能的发展大体呈现出若干可以预见的趋势。在方法层面,如何将擅长感知与归纳的连接主义方法,同擅长推理与可解释的符号主义方法结合起来,即所谓神经符号融合,被普遍视为通向更强、更可信智能的重要路径,这也意味着经典理论并未过时,而是有望以新的形式重新融入前沿研究。在能力层面,让智能体具备自主规划、使用工具、与环境持续交互并完成复杂任务的\"智能体\"方向正受到越来越多的重视,人工智能正从被动应答走向主动执行。在效率层面,模型的轻量化、蒸馏与边缘部署将使智能能力从云端走向终端,降低使用门槛。在可信层面,提升可解释性、安全性与价值对齐,建立与之配套的评测标准和治理框架,将与能力提升同等重要。至于是否以及何时能够实现具备广泛迁移能力、接近人类通用智能的通用人工智能,学界仍存在很大分歧,它更应被视为一个牵引研究方向的长远愿景,而非近在眼前的目标。",[1095,1539,1540],{},"最后对于学习者而言,唯有既扎实掌握经典理论、理解其思想脉络,又保持对前沿进展和现实问题的敏感,才能在这个充满机遇与挑战的领域中形成全面而理性的认识。",{"title":1542,"searchDepth":1543,"depth":1543,"links":1544},"",4,[],"（备课用）人工智能导论（人工智能相关串讲）","md",true,{"uuid":1549,"slots":1550},"82044f80-eb35-11f0-9469-dddde1da8d6f",{},38,{"title":1085,"description":1545},"posts\u002F2026\u002F2026-01-06-人工智能导论",[236],"c8z6QYJoAcGGRiQ-ZS28GrGqC7whLFqB-gH1wfGAD0Y",1790443286567]