[{"data":1,"prerenderedAt":3367},["ShallowReactive",2],{"post-arxiv-2005-03448v3":3,"site-info":2662,"related-en-arxiv-2005-03448v3":2758},{"zh":4,"en":1414},{"id":5,"title":6,"author":7,"body":8,"cover":123,"coverAlt":6,"date":1397,"description":1398,"draft":1399,"extension":1400,"featured":1399,"lang":1401,"meta":1402,"navigation":1403,"path":1404,"seo":1405,"slug":1406,"stem":1407,"tags":1408,"translationKey":1406,"updated":1412,"__hash__":1413},"blog\u002Fblog\u002Fzh-TW\u002Farxiv-2005-03448v3.md","基於稀缺資料的物理資訊學習：從資料中發現控制方程式","Jecision",{"type":9,"value":10,"toc":1379},"minimark",[11,22,27,30,33,68,86,89,92,105,108,114,117,124,128,137,141,147,881,886,897,902,916,921,929,933,1012,1017,1025,1031,1360,1364],[12,13,15,16],"h2",{"id":14},"論文位置httpsarxivorgabs200503448","論文位置：",[17,18,19],"a",{"href":19,"rel":20},"https:\u002F\u002Farxiv.org\u002Fabs\u002F2005.03448",[21],"nofollow",[23,24,26],"h3",{"id":25},"physics-informed-learning-of-governing-equations-from-scarce-data","Physics-informed learning of governing equations from scarce data",[23,28,29],{"id":29},"內容簡易翻譯",[12,31,32],{"id":32},"作者",[34,35,36,40,44,45,48,51,52,55,58,60,61,63,64,67],"p",{},[37,38,39],"strong",{},"陳鑷（Zhao Chen）",[41,42,43],"sup",{},"1","，*",[37,46,47],{},"劉陽（Yang Liu）",[41,49,50],{},"2","，",[37,53,54],{},"孫浩（Hao Sun）",[41,56,57],{},"1,3",[41,59,43],{}," 美國東北大學土木與環境工程系（波士頓）\n",[41,62,50],{}," 美國東北大學機械與工業工程系（波士頓）\n",[41,65,66],{},"3"," 麻省理工學院土木與環境工程系（劍橋）",[69,70,71],"blockquote",{},[34,72,73],{},[74,75,76,77,81,82],"em",{},"通訊作者信箱: ",[17,78,80],{"href":79},"mailto:yang1.liu@northeastern.edu","yang1.liu@northeastern.edu","; ",[17,83,85],{"href":84},"mailto:h.sun@northeastern.edu","h.sun@northeastern.edu",[12,87,88],{"id":88},"摘要",[34,90,91],{},"利用數據揭示描述複雜物理系統行為的基本控制定律或方程，能顯著推動多學科領域的建模、模擬與理解。本文提出一種新型物理資訊深度學習架構（PiDL），用於從稀缺且含雜訊的資料中發現非線性時空系統的控制偏微分方程（PDE）。此方法深度融合了深度神經網路（DNN）的表示學習能力、物理嵌入、自動微分與稀疏迴歸技術，實現三大目標：",[93,94,95,99,102],"ol",{},[96,97,98],"li",{},"逼近系統變數解",[96,100,101],{},"計算關鍵導數項",[96,103,104],{},"辨識構成PDE結構與明確表達的關鍵導數項及參數",[34,106,107],{},"透過數值與實驗驗證，此方法在資料稀缺性和雜訊幹擾下對多種PDE系統（考慮不同初\u002F邊值條件）的發現具有高效性與穩健性。該框架為實際應用中難以獲取大規模精確資料集的問題提供了閉式模型發現的潛力。",[34,109,110,113],{},[37,111,112],{},"關鍵字","：控制方程式發現、物理資訊深度學習、偏微分方程（PDE）",[12,115,116],{"id":116},"核心貢獻與重點",[34,118,119],{},[120,121],"img",{"alt":122,"src":123},"image","\u002Fimages\u002Fblog\u002F2005.03448v3.webp",[23,125,127],{"id":126},"_1-問題背景","1. 問題背景",[129,130,131,134],"ul",{},[96,132,133],{},"傳統PDE建模依賴第一原理（如守恆定律），但許多複雜系統的解析描述未知",[96,135,136],{},"資料驅動方法（如SINDy）受限於資料品質與數量，數值微分對雜訊敏感",[23,138,140],{"id":139},"_2-pidl框架創新","2. PiDL框架創新",[34,142,143,146],{},[37,144,145],{},"損失函數","：",[148,149,152],"span",{"className":150},[151],"katex-display",[148,153,156,273],{"className":154},[155],"katex",[148,157,160],{"className":158},[159],"katex-mathml",[161,162,165],"math",{"xmlns":163,"display":164},"http:\u002F\u002Fwww.w3.org\u002F1998\u002FMath\u002FMathML","block",[166,167,168,268],"semantics",{},[169,170,171,176,181,184,188,192,195,198,218,221,224,239,241,244],"mrow",{},[172,173,175],"mi",{"mathvariant":174},"script","L",[177,178,180],"mo",{"stretchy":179},"false","(",[172,182,183],{},"θ",[177,185,187],{"separator":186},"true",",",[172,189,191],{"mathvariant":190},"normal","Λ",[177,193,194],{"stretchy":179},")",[177,196,197],{},"=",[199,200,201,214],"munder",{},[199,202,203,211],{},[204,205,206,208],"msub",{},[172,207,175],{"mathvariant":174},[172,209,210],{},"d",[177,212,213],{"stretchy":186},"⏟",[215,216,217],"mtext",{},"數據損失",[177,219,220],{},"+",[172,222,223],{},"α",[199,225,226,236],{},[199,227,228,234],{},[204,229,230,232],{},[172,231,175],{"mathvariant":174},[172,233,34],{},[177,235,213],{"stretchy":186},[215,237,238],{},"物理損失",[177,240,220],{},[172,242,243],{},"β",[199,245,246,265],{},[199,247,248,263],{},[169,249,250,253,255],{},[172,251,252],{"mathvariant":190},"∥",[172,254,191],{"mathvariant":190},[204,256,257,259],{},[172,258,252],{"mathvariant":190},[260,261,262],"mn",{},"0",[177,264,213],{"stretchy":186},[215,266,267],{},"稀疏正則",[269,270,272],"annotation",{"encoding":271},"application\u002Fx-tex","\\mathcal{L}(\\theta, \\Lambda) = \\underbrace{\\mathcal{L}_d}_{\\text{數據損失}} + \\alpha \\underbrace{\\mathcal{L}_p}_{\\text{物理損失}} + \\beta \\underbrace{\\|\\Lambda\\|_0}_{\\text{稀疏正則}}",[148,274,277,327,536,711],{"className":275,"ariaHidden":186},[276],"katex-html",[148,278,281,286,291,295,300,304,309,312,316,320,324],{"className":279},[280],"base",[148,282],{"className":283,"style":285},[284],"strut","height:1em;vertical-align:-0.25em;",[148,287,175],{"className":288},[289,290],"mord","mathcal",[148,292,180],{"className":293},[294],"mopen",[148,296,183],{"className":297,"style":299},[289,298],"mathnormal","margin-right:0.0278em;",[148,301,187],{"className":302},[303],"mpunct",[148,305],{"className":306,"style":308},[307],"mspace","margin-right:0.1667em;",[148,310,191],{"className":311},[289],[148,313,194],{"className":314},[315],"mclose",[148,317],{"className":318,"style":319},[307],"margin-right:0.2778em;",[148,321,197],{"className":322},[323],"mrel",[148,325],{"className":326,"style":319},[307],[148,328,330,334,525,529,533],{"className":329},[280],[148,331],{"className":332,"style":333},[284],"height:2.1597em;vertical-align:-1.4763em;",[148,335,338],{"className":336},[337,199],"minner",[148,339,343,516],{"className":340},[341,342],"vlist-t","vlist-t2",[148,344,347,513],{"className":345},[346],"vlist-r",[148,348,352,378],{"className":349,"style":351},[350],"vlist","height:0.6833em;",[148,353,355,360],{"style":354},"top:-1.5237em;",[148,356],{"className":357,"style":359},[358],"pstrut","height:3em;",[148,361,367],{"className":362},[363,364,365,366],"sizing","reset-size6","size3","mtight",[148,368,370],{"className":369},[289,366],[148,371,374],{"className":372},[289,373,366],"text",[148,375,217],{"className":376},[289,377,366],"cjk_fallback",[148,379,381,384],{"style":380},"top:-3em;",[148,382],{"className":383,"style":359},[358],[148,385,387],{"className":386},[337,199],[148,388,390,504],{"className":389},[341,342],[148,391,393,501],{"className":392},[346],[148,394,396,446],{"className":395,"style":351},[350],[148,397,401,404],{"className":398,"style":400},[399],"svg-align","top:-2.202em;",[148,402],{"className":403,"style":359},[358],[148,405,409,426,436],{"className":406,"style":408},[407],"stretchy","height:0.548em;min-width:1.6em;",[148,410,414],{"className":411,"style":413},[412],"brace-left","height:0.548em;",[415,416,422],"svg",{"xmlns":417,"width":418,"height":419,"viewBox":420,"preserveAspectRatio":421},"http:\u002F\u002Fwww.w3.org\u002F2000\u002Fsvg","400em","0.548em","0 0 400000 548","xMinYMin slice",[423,424],"path",{"d":425},"M0 6l6-6h17c12.688 0 19.313.3 20 1 4 4 7.313 8.3 10 13\n 35.313 51.3 80.813 93.8 136.5 127.5 55.688 33.7 117.188 55.8 184.5 66.5.688\n 0 2 .3 4 1 18.688 2.7 76 4.3 172 5h399450v120H429l-6-1c-124.688-8-235-61.7\n-331-161C60.687 138.7 32.312 99.3 7 54L0 41V6z",[148,427,430],{"className":428,"style":413},[429],"brace-center",[415,431,433],{"xmlns":417,"width":418,"height":419,"viewBox":420,"preserveAspectRatio":432},"xMidYMin slice",[423,434],{"d":435},"M199572 214\nc100.7 8.3 195.3 44 280 108 55.3 42 101.7 93 139 153l9 14c2.7-4 5.7-8.7 9-14\n 53.3-86.7 123.7-153 211-199 66.7-36 137.3-56.3 212-62h199568v120H200432c-178.3\n 11.7-311.7 78.3-403 201-6 8-9.7 12-11 12-.7.7-6.7 1-18 1s-17.3-.3-18-1c-1.3 0\n-5-4-11-12-44.7-59.3-101.3-106.3-170-141s-145.3-54.3-229-60H0V214z",[148,437,440],{"className":438,"style":413},[439],"brace-right",[415,441,443],{"xmlns":417,"width":418,"height":419,"viewBox":420,"preserveAspectRatio":442},"xMaxYMin slice",[423,444],{"d":445},"M399994 0l6 6v35l-6 11c-56 104-135.3 181.3-238 232-57.3\n 28.7-117 45-179 50H-300V214h399897c43.3-7 81-15 113-26 100.7-33 179.7-91 237\n-174 2.7-5 6-9 10-13 .7-1 7.3-1 20-1h17z",[148,447,448,451],{"style":380},[148,449],{"className":450,"style":359},[358],[148,452,454],{"className":453},[289],[148,455,457,460],{"className":456},[289],[148,458,175],{"className":459},[289,290],[148,461,464],{"className":462},[463],"msupsub",[148,465,467,492],{"className":466},[341,342],[148,468,470,487],{"className":469},[346],[148,471,474],{"className":472,"style":473},[350],"height:0.3361em;",[148,475,477,481],{"style":476},"top:-2.55em;margin-left:0em;margin-right:0.05em;",[148,478],{"className":479,"style":480},[358],"height:2.7em;",[148,482,484],{"className":483},[363,364,365,366],[148,485,210],{"className":486},[289,298,366],[148,488,491],{"className":489},[490],"vlist-s","​",[148,493,495],{"className":494},[346],[148,496,499],{"className":497,"style":498},[350],"height:0.15em;",[148,500],{},[148,502,491],{"className":503},[490],[148,505,507],{"className":506},[346],[148,508,511],{"className":509,"style":510},[350],"height:0.798em;",[148,512],{},[148,514,491],{"className":515},[490],[148,517,519],{"className":518},[346],[148,520,523],{"className":521,"style":522},[350],"height:1.4763em;",[148,524],{},[148,526],{"className":527,"style":528},[307],"margin-right:0.2222em;",[148,530,220],{"className":531},[532],"mbin",[148,534],{"className":535,"style":528},[307],[148,537,539,543,547,550,702,705,708],{"className":538},[280],[148,540],{"className":541,"style":542},[284],"height:2.2958em;vertical-align:-1.6124em;",[148,544,223],{"className":545,"style":546},[289,298],"margin-right:0.0037em;",[148,548],{"className":549,"style":308},[307],[148,551,553],{"className":552},[337,199],[148,554,556,693],{"className":555},[341,342],[148,557,559,690],{"className":558},[346],[148,560,562,580],{"className":561,"style":351},[350],[148,563,565,568],{"style":564},"top:-1.3876em;",[148,566],{"className":567,"style":359},[358],[148,569,571],{"className":570},[363,364,365,366],[148,572,574],{"className":573},[289,366],[148,575,577],{"className":576},[289,373,366],[148,578,238],{"className":579},[289,377,366],[148,581,582,585],{"style":380},[148,583],{"className":584,"style":359},[358],[148,586,588],{"className":587},[337,199],[148,589,591,681],{"className":590},[341,342],[148,592,594,678],{"className":593},[346],[148,595,597,628],{"className":596,"style":351},[350],[148,598,601,604],{"className":599,"style":600},[399],"top:-2.0659em;",[148,602],{"className":603,"style":359},[358],[148,605,607,614,621],{"className":606,"style":408},[407],[148,608,610],{"className":609,"style":413},[412],[415,611,612],{"xmlns":417,"width":418,"height":419,"viewBox":420,"preserveAspectRatio":421},[423,613],{"d":425},[148,615,617],{"className":616,"style":413},[429],[415,618,619],{"xmlns":417,"width":418,"height":419,"viewBox":420,"preserveAspectRatio":432},[423,620],{"d":435},[148,622,624],{"className":623,"style":413},[439],[415,625,626],{"xmlns":417,"width":418,"height":419,"viewBox":420,"preserveAspectRatio":442},[423,627],{"d":445},[148,629,630,633],{"style":380},[148,631],{"className":632,"style":359},[358],[148,634,636],{"className":635},[289],[148,637,639,642],{"className":638},[289],[148,640,175],{"className":641},[289,290],[148,643,645],{"className":644},[463],[148,646,648,669],{"className":647},[341,342],[148,649,651,666],{"className":650},[346],[148,652,655],{"className":653,"style":654},[350],"height:0.1514em;",[148,656,657,660],{"style":476},[148,658],{"className":659,"style":480},[358],[148,661,663],{"className":662},[363,364,365,366],[148,664,34],{"className":665},[289,298,366],[148,667,491],{"className":668},[490],[148,670,672],{"className":671},[346],[148,673,676],{"className":674,"style":675},[350],"height:0.2861em;",[148,677],{},[148,679,491],{"className":680},[490],[148,682,684],{"className":683},[346],[148,685,688],{"className":686,"style":687},[350],"height:0.9341em;",[148,689],{},[148,691,491],{"className":692},[490],[148,694,696],{"className":695},[346],[148,697,700],{"className":698,"style":699},[350],"height:1.6124em;",[148,701],{},[148,703],{"className":704,"style":528},[307],[148,706,220],{"className":707},[532],[148,709],{"className":710,"style":528},[307],[148,712,714,718,722,725],{"className":713},[280],[148,715],{"className":716,"style":717},[284],"height:2.3263em;vertical-align:-1.5763em;",[148,719,243],{"className":720,"style":721},[289,298],"margin-right:0.0528em;",[148,723],{"className":724,"style":308},[307],[148,726,728],{"className":727},[337,199],[148,729,731,872],{"className":730},[341,342],[148,732,734,869],{"className":733},[346],[148,735,738,756],{"className":736,"style":737},[350],"height:0.75em;",[148,739,741,744],{"style":740},"top:-1.4237em;",[148,742],{"className":743,"style":359},[358],[148,745,747],{"className":746},[363,364,365,366],[148,748,750],{"className":749},[289,366],[148,751,753],{"className":752},[289,373,366],[148,754,267],{"className":755},[289,377,366],[148,757,758,761],{"style":380},[148,759],{"className":760,"style":359},[358],[148,762,764],{"className":763},[337,199],[148,765,767,860],{"className":766},[341,342],[148,768,770,857],{"className":769},[346],[148,771,773,804],{"className":772,"style":737},[350],[148,774,777,780],{"className":775,"style":776},[399],"top:-2.102em;",[148,778],{"className":779,"style":359},[358],[148,781,783,790,797],{"className":782,"style":408},[407],[148,784,786],{"className":785,"style":413},[412],[415,787,788],{"xmlns":417,"width":418,"height":419,"viewBox":420,"preserveAspectRatio":421},[423,789],{"d":425},[148,791,793],{"className":792,"style":413},[429],[415,794,795],{"xmlns":417,"width":418,"height":419,"viewBox":420,"preserveAspectRatio":432},[423,796],{"d":435},[148,798,800],{"className":799,"style":413},[439],[415,801,802],{"xmlns":417,"width":418,"height":419,"viewBox":420,"preserveAspectRatio":442},[423,803],{"d":445},[148,805,806,809],{"style":380},[148,807],{"className":808,"style":359},[358],[148,810,812,816],{"className":811},[289],[148,813,815],{"className":814},[289],"∥Λ",[148,817,819,822],{"className":818},[289],[148,820,252],{"className":821},[289],[148,823,825],{"className":824},[463],[148,826,828,849],{"className":827},[341,342],[148,829,831,846],{"className":830},[346],[148,832,835],{"className":833,"style":834},[350],"height:0.3011em;",[148,836,837,840],{"style":476},[148,838],{"className":839,"style":480},[358],[148,841,843],{"className":842},[363,364,365,366],[148,844,262],{"className":845},[289,366],[148,847,491],{"className":848},[490],[148,850,852],{"className":851},[346],[148,853,855],{"className":854,"style":498},[350],[148,856],{},[148,858,491],{"className":859},[490],[148,861,863],{"className":862},[346],[148,864,867],{"className":865,"style":866},[350],"height:0.898em;",[148,868],{},[148,870,491],{"className":871},[490],[148,873,875],{"className":874},[346],[148,876,879],{"className":877,"style":878},[350],"height:1.5763em;",[148,880],{},[34,882,883,146],{},[37,884,885],{},"技術融合",[129,887,888,891,894],{},[96,889,890],{},"DNN學習系統解的表示",[96,892,893],{},"自動微分高精度計算導數",[96,895,896],{},"ℓ₀稀疏迴歸辨識PDE關鍵項",[34,898,899,146],{},[37,900,901],{},"網路架構",[129,903,904,910],{},[96,905,906,909],{},[37,907,908],{},"單一資料集處理","：\n└─ DNN直接建模解函數",[96,911,912,915],{},[37,913,914],{},"多重資料集處理","：\n└─ \"根-分支\"DNN結構處理不同初邊值條件\n    ├─ 共享根網路擷取共同特徵\n    └─ 獨立分支網路適應特定條件",[34,917,918,146],{},[37,919,920],{},"最佳化演算法（ADO）",[129,922,923,926],{},[96,924,925],{},"交替方向優化逐步訓練DNN參數(θ)與PDE係數(Λ)",[96,927,928],{},"解決ℓ₀正規的NP難問題",[23,930,932],{"id":931},"_3-實驗驗證","3. 實驗驗證",[934,935,936,955],"table",{},[937,938,939],"thead",{},[940,941,942,946,949,952],"tr",{},[943,944,945],"th",{},"PDE類型",[943,947,948],{},"雜訊等級",[943,950,951],{},"辨識誤差",[943,953,954],{},"資料使用量",[956,957,958,973,986,999],"tbody",{},[940,959,960,964,967,970],{},[961,962,963],"td",{},"Burgers方程式",[961,965,966],{},"10%",[961,968,969],{},"1.15%",[961,971,972],{},"1.95%",[940,974,975,978,980,983],{},[961,976,977],{},"KS方程式（混沌）",[961,979,966],{},[961,981,982],{},"0.71%",[961,984,985],{},"12.3%",[940,987,988,991,993,996],{},[961,989,990],{},"Navier-Stokes方程式",[961,992,966],{},[961,994,995],{},"1.40%",[961,997,998],{},"0.22%",[940,1000,1001,1004,1006,1009],{},[961,1002,1003],{},"Schrödinger方程式",[961,1005,966],{},[961,1007,1008],{},"2.31%",[961,1010,1011],{},"37.5%",[34,1013,1014,146],{},[37,1015,1016],{},"多重資料集場景",[129,1018,1019,1022],{},[96,1020,1021],{},"Burgers方程式（小黏度衝擊波）",[96,1023,1024],{},"Fitzhugh-Nagumo神經元模型",[34,1026,1027,1030],{},[37,1028,1029],{},"真實實驗","：\n細胞遷移增殖實驗發現Fisher-Kolmogorov模型：",[148,1032,1034],{"className":1033},[151],[148,1035,1037,1099],{"className":1036},[155],[148,1038,1040],{"className":1039},[159],[161,1041,1042],{"xmlns":163,"display":164},[166,1043,1044,1096],{},[169,1045,1046,1054,1056,1059,1070,1072,1079,1081,1083,1089],{},[204,1047,1048,1051],{},[172,1049,1050],{},"ρ",[172,1052,1053],{},"t",[177,1055,197],{},[172,1057,1058],{},"γ",[204,1060,1061,1063],{},[172,1062,1050],{},[169,1064,1065,1068],{},[172,1066,1067],{},"x",[172,1069,1067],{},[177,1071,220],{},[204,1073,1074,1077],{},[172,1075,1076],{},"λ",[260,1078,43],{},[172,1080,1050],{},[177,1082,220],{},[204,1084,1085,1087],{},[172,1086,1076],{},[260,1088,50],{},[1090,1091,1092,1094],"msup",{},[172,1093,1050],{},[260,1095,50],{},[269,1097,1098],{"encoding":271},"\\rho_t = \\gamma \\rho_{xx} + \\lambda_1 \\rho + \\lambda_2 \\rho^2",[148,1100,1102,1159,1223,1282],{"className":1101,"ariaHidden":186},[276],[148,1103,1105,1109,1150,1153,1156],{"className":1104},[280],[148,1106],{"className":1107,"style":1108},[284],"height:0.625em;vertical-align:-0.1944em;",[148,1110,1112,1115],{"className":1111},[289],[148,1113,1050],{"className":1114},[289,298],[148,1116,1118],{"className":1117},[463],[148,1119,1121,1142],{"className":1120},[341,342],[148,1122,1124,1139],{"className":1123},[346],[148,1125,1128],{"className":1126,"style":1127},[350],"height:0.2806em;",[148,1129,1130,1133],{"style":476},[148,1131],{"className":1132,"style":480},[358],[148,1134,1136],{"className":1135},[363,364,365,366],[148,1137,1053],{"className":1138},[289,298,366],[148,1140,491],{"className":1141},[490],[148,1143,1145],{"className":1144},[346],[148,1146,1148],{"className":1147,"style":498},[350],[148,1149],{},[148,1151],{"className":1152,"style":319},[307],[148,1154,197],{"className":1155},[323],[148,1157],{"className":1158,"style":319},[307],[148,1160,1162,1166,1170,1214,1217,1220],{"className":1161},[280],[148,1163],{"className":1164,"style":1165},[284],"height:0.7778em;vertical-align:-0.1944em;",[148,1167,1058],{"className":1168,"style":1169},[289,298],"margin-right:0.0556em;",[148,1171,1173,1176],{"className":1172},[289],[148,1174,1050],{"className":1175},[289,298],[148,1177,1179],{"className":1178},[463],[148,1180,1182,1206],{"className":1181},[341,342],[148,1183,1185,1203],{"className":1184},[346],[148,1186,1188],{"className":1187,"style":654},[350],[148,1189,1190,1193],{"style":476},[148,1191],{"className":1192,"style":480},[358],[148,1194,1196],{"className":1195},[363,364,365,366],[148,1197,1199],{"className":1198},[289,366],[148,1200,1202],{"className":1201},[289,298,366],"xx",[148,1204,491],{"className":1205},[490],[148,1207,1209],{"className":1208},[346],[148,1210,1212],{"className":1211,"style":498},[350],[148,1213],{},[148,1215],{"className":1216,"style":528},[307],[148,1218,220],{"className":1219},[532],[148,1221],{"className":1222,"style":528},[307],[148,1224,1226,1230,1270,1273,1276,1279],{"className":1225},[280],[148,1227],{"className":1228,"style":1229},[284],"height:0.8889em;vertical-align:-0.1944em;",[148,1231,1233,1236],{"className":1232},[289],[148,1234,1076],{"className":1235},[289,298],[148,1237,1239],{"className":1238},[463],[148,1240,1242,1262],{"className":1241},[341,342],[148,1243,1245,1259],{"className":1244},[346],[148,1246,1248],{"className":1247,"style":834},[350],[148,1249,1250,1253],{"style":476},[148,1251],{"className":1252,"style":480},[358],[148,1254,1256],{"className":1255},[363,364,365,366],[148,1257,43],{"className":1258},[289,366],[148,1260,491],{"className":1261},[490],[148,1263,1265],{"className":1264},[346],[148,1266,1268],{"className":1267,"style":498},[350],[148,1269],{},[148,1271,1050],{"className":1272},[289,298],[148,1274],{"className":1275,"style":528},[307],[148,1277,220],{"className":1278},[532],[148,1280],{"className":1281,"style":528},[307],[148,1283,1285,1289,1329],{"className":1284},[280],[148,1286],{"className":1287,"style":1288},[284],"height:1.0585em;vertical-align:-0.1944em;",[148,1290,1292,1295],{"className":1291},[289],[148,1293,1076],{"className":1294},[289,298],[148,1296,1298],{"className":1297},[463],[148,1299,1301,1321],{"className":1300},[341,342],[148,1302,1304,1318],{"className":1303},[346],[148,1305,1307],{"className":1306,"style":834},[350],[148,1308,1309,1312],{"style":476},[148,1310],{"className":1311,"style":480},[358],[148,1313,1315],{"className":1314},[363,364,365,366],[148,1316,50],{"className":1317},[289,366],[148,1319,491],{"className":1320},[490],[148,1322,1324],{"className":1323},[346],[148,1325,1327],{"className":1326,"style":498},[350],[148,1328],{},[148,1330,1332,1335],{"className":1331},[289],[148,1333,1050],{"className":1334},[289,298],[148,1336,1338],{"className":1337},[463],[148,1339,1341],{"className":1340},[341],[148,1342,1344],{"className":1343},[346],[148,1345,1348],{"className":1346,"style":1347},[350],"height:0.8641em;",[148,1349,1351,1354],{"style":1350},"top:-3.113em;margin-right:0.05em;",[148,1352],{"className":1353,"style":480},[358],[148,1355,1357],{"className":1356},[363,364,365,366],[148,1358,50],{"className":1359},[289,366],[23,1361,1363],{"id":1362},"_4-優勢與意義","4. 優勢與意義",[34,1365,1366,1367,1370,1371,1374,1375,1378],{},"✅ ",[37,1368,1369],{},"突破限制","：資料稀缺(\u003C1%原網格)+高雜訊(10%)下精確發現PDE\n✅ ",[37,1372,1373],{},"物理可解釋性","：DNN提供全域逼近，稀疏PDE約束網絡\n✅ ",[37,1376,1377],{},"應用潛力","：流體力學\u002F生物遷移\u002F氣候預測等高成本實測領域",{"title":1380,"searchDepth":1381,"depth":1381,"links":1382},"",2,[1383,1389,1390,1391],{"id":14,"depth":1381,"text":1384,"children":1385},"論文位置：https:\u002F\u002Farxiv.org\u002Fabs\u002F2005.03448",[1386,1388],{"id":25,"depth":1387,"text":26},3,{"id":29,"depth":1387,"text":29},{"id":32,"depth":1381,"text":32},{"id":88,"depth":1381,"text":88},{"id":116,"depth":1381,"text":116,"children":1392},[1393,1394,1395,1396],{"id":126,"depth":1387,"text":127},{"id":139,"depth":1387,"text":140},{"id":931,"depth":1387,"text":932},{"id":1362,"depth":1387,"text":1363},"2025-07-09","PiDL 用稀疏又帶噪音的觀測資料反推非線性時空系統的偏微分方程；本文整理它的網路架構、訓練策略，以及在 Burgers 與 Navier–Stokes 上的實測表現。",false,"md","zh-TW",{},true,"\u002Fblog\u002Fzh-tw\u002Farxiv-2005-03448v3",{"title":6,"description":1398},"arxiv-2005-03448v3","blog\u002Fzh-TW\u002Farxiv-2005-03448v3",[1409,1410,1411],"paper","physics-informed","pde",null,"_lKytP4yva134MGngrv1HaVh0_rypK39ea4Sq0NPTS4",{"id":1415,"title":1416,"author":7,"body":1417,"cover":123,"coverAlt":26,"date":1397,"description":2654,"draft":1399,"extension":1400,"featured":1399,"lang":2655,"meta":2656,"navigation":1403,"path":2657,"seo":2658,"slug":1406,"stem":2659,"tags":2660,"translationKey":1406,"updated":1412,"__hash__":2661},"blog\u002Fblog\u002Fen\u002Farxiv-2005-03448v3.md","Discovering governing equations from scarce data with PiDL",{"type":9,"value":1418,"toc":2646},[1419,1437,1441,1448,1451,1462,1468,1472,1475,1478,1482,1485,2113,2118,2129,2134,2156,2223,2227,2291,2294,2297,2610,2614,2620,2626,2632,2636,2643],[34,1420,1421,1424,1425,1429,1432,1433,1436],{},[37,1422,1423],{},"Paper"," — ",[17,1426,1428],{"href":19,"rel":1427},[21],"arXiv:2005.03448",[37,1430,1431],{},"Title"," — Physics-informed learning of governing equations from scarce data\n",[37,1434,1435],{},"Authors"," — Zhao Chen¹, Yang Liu², Hao Sun¹˒³\n¹ Northeastern University, Civil & Environmental Engineering\n² Northeastern University, Mechanical & Industrial Engineering\n³ MIT, Civil & Environmental Engineering",[12,1438,1440],{"id":1439},"abstract-in-short","Abstract, in short",[34,1442,1443,1444,1447],{},"Recovering the governing equations of a complex physical system from data is what turns observations into a model. This paper proposes PiDL, a physics-informed deep learning architecture that discovers the governing PDEs of nonlinear spatio-temporal systems from ",[37,1445,1446],{},"scarce and noisy"," data.",[34,1449,1450],{},"It fuses four things: the representation power of deep neural networks, physical embedding, automatic differentiation, and sparse regression. Three goals:",[93,1452,1453,1456,1459],{},[96,1454,1455],{},"Approximate the solution of the system state",[96,1457,1458],{},"Compute the relevant derivative terms",[96,1460,1461],{},"Identify which terms and coefficients actually form the PDE",[34,1463,1464,1467],{},[37,1465,1466],{},"Keywords"," — governing equation discovery, physics-informed deep learning, PDEs",[12,1469,1471],{"id":1470},"why-the-problem-is-hard","Why the problem is hard",[34,1473,1474],{},"Classical PDE modelling relies on first principles — conservation laws and the like. But for many complex systems, the analytical form is simply unknown.",[34,1476,1477],{},"Data-driven methods such as SINDy are limited by data quality and quantity, and numerical differentiation is notoriously sensitive to noise. Ask for derivatives from noisy, sparse measurements and the noise gets amplified into the coefficients you are trying to identify.",[12,1479,1481],{"id":1480},"the-pidl-framework","The PiDL framework",[34,1483,1484],{},"The loss combines data fit, physics residual, and sparsity:",[148,1486,1488],{"className":1487},[151],[148,1489,1491,1576],{"className":1490},[155],[148,1492,1494],{"className":1493},[159],[161,1495,1496],{"xmlns":163,"display":164},[166,1497,1498,1573],{},[169,1499,1500,1502,1504,1506,1508,1510,1512,1514,1529,1531,1533,1548,1550,1552],{},[172,1501,175],{"mathvariant":174},[177,1503,180],{"stretchy":179},[172,1505,183],{},[177,1507,187],{"separator":186},[172,1509,191],{"mathvariant":190},[177,1511,194],{"stretchy":179},[177,1513,197],{},[199,1515,1516,1526],{},[199,1517,1518,1524],{},[204,1519,1520,1522],{},[172,1521,175],{"mathvariant":174},[172,1523,210],{},[177,1525,213],{"stretchy":186},[215,1527,1528],{},"data",[177,1530,220],{},[172,1532,223],{},[199,1534,1535,1545],{},[199,1536,1537,1543],{},[204,1538,1539,1541],{},[172,1540,175],{"mathvariant":174},[172,1542,34],{},[177,1544,213],{"stretchy":186},[215,1546,1547],{},"physics",[177,1549,220],{},[172,1551,243],{},[199,1553,1554,1570],{},[199,1555,1556,1568],{},[169,1557,1558,1560,1562],{},[172,1559,252],{"mathvariant":190},[172,1561,191],{"mathvariant":190},[204,1563,1564,1566],{},[172,1565,252],{"mathvariant":190},[260,1567,262],{},[177,1569,213],{"stretchy":186},[215,1571,1572],{},"sparsity",[269,1574,1575],{"encoding":271},"\\mathcal{L}(\\theta, \\Lambda) = \\underbrace{\\mathcal{L}_d}_{\\text{data}} + \\alpha \\underbrace{\\mathcal{L}_p}_{\\text{physics}} + \\beta \\underbrace{\\|\\Lambda\\|_0}_{\\text{sparsity}}",[148,1577,1579,1615,1779,1949],{"className":1578,"ariaHidden":186},[276],[148,1580,1582,1585,1588,1591,1594,1597,1600,1603,1606,1609,1612],{"className":1581},[280],[148,1583],{"className":1584,"style":285},[284],[148,1586,175],{"className":1587},[289,290],[148,1589,180],{"className":1590},[294],[148,1592,183],{"className":1593,"style":299},[289,298],[148,1595,187],{"className":1596},[303],[148,1598],{"className":1599,"style":308},[307],[148,1601,191],{"className":1602},[289],[148,1604,194],{"className":1605},[315],[148,1607],{"className":1608,"style":319},[307],[148,1610,197],{"className":1611},[323],[148,1613],{"className":1614,"style":319},[307],[148,1616,1618,1622,1770,1773,1776],{"className":1617},[280],[148,1619],{"className":1620,"style":1621},[284],"height:2.1674em;vertical-align:-1.4841em;",[148,1623,1625],{"className":1624},[337,199],[148,1626,1628,1761],{"className":1627},[341,342],[148,1629,1631,1758],{"className":1630},[346],[148,1632,1634,1652],{"className":1633,"style":351},[350],[148,1635,1637,1640],{"style":1636},"top:-1.5159em;",[148,1638],{"className":1639,"style":359},[358],[148,1641,1643],{"className":1642},[363,364,365,366],[148,1644,1646],{"className":1645},[289,366],[148,1647,1649],{"className":1648},[289,373,366],[148,1650,1528],{"className":1651},[289,366],[148,1653,1654,1657],{"style":380},[148,1655],{"className":1656,"style":359},[358],[148,1658,1660],{"className":1659},[337,199],[148,1661,1663,1750],{"className":1662},[341,342],[148,1664,1666,1747],{"className":1665},[346],[148,1667,1669,1699],{"className":1668,"style":351},[350],[148,1670,1672,1675],{"className":1671,"style":400},[399],[148,1673],{"className":1674,"style":359},[358],[148,1676,1678,1685,1692],{"className":1677,"style":408},[407],[148,1679,1681],{"className":1680,"style":413},[412],[415,1682,1683],{"xmlns":417,"width":418,"height":419,"viewBox":420,"preserveAspectRatio":421},[423,1684],{"d":425},[148,1686,1688],{"className":1687,"style":413},[429],[415,1689,1690],{"xmlns":417,"width":418,"height":419,"viewBox":420,"preserveAspectRatio":432},[423,1691],{"d":435},[148,1693,1695],{"className":1694,"style":413},[439],[415,1696,1697],{"xmlns":417,"width":418,"height":419,"viewBox":420,"preserveAspectRatio":442},[423,1698],{"d":445},[148,1700,1701,1704],{"style":380},[148,1702],{"className":1703,"style":359},[358],[148,1705,1707],{"className":1706},[289],[148,1708,1710,1713],{"className":1709},[289],[148,1711,175],{"className":1712},[289,290],[148,1714,1716],{"className":1715},[463],[148,1717,1719,1739],{"className":1718},[341,342],[148,1720,1722,1736],{"className":1721},[346],[148,1723,1725],{"className":1724,"style":473},[350],[148,1726,1727,1730],{"style":476},[148,1728],{"className":1729,"style":480},[358],[148,1731,1733],{"className":1732},[363,364,365,366],[148,1734,210],{"className":1735},[289,298,366],[148,1737,491],{"className":1738},[490],[148,1740,1742],{"className":1741},[346],[148,1743,1745],{"className":1744,"style":498},[350],[148,1746],{},[148,1748,491],{"className":1749},[490],[148,1751,1753],{"className":1752},[346],[148,1754,1756],{"className":1755,"style":510},[350],[148,1757],{},[148,1759,491],{"className":1760},[490],[148,1762,1764],{"className":1763},[346],[148,1765,1768],{"className":1766,"style":1767},[350],"height:1.4841em;",[148,1769],{},[148,1771],{"className":1772,"style":528},[307],[148,1774,220],{"className":1775},[532],[148,1777],{"className":1778,"style":528},[307],[148,1780,1782,1786,1789,1792,1940,1943,1946],{"className":1781},[280],[148,1783],{"className":1784,"style":1785},[284],"height:2.4397em;vertical-align:-1.7563em;",[148,1787,223],{"className":1788,"style":546},[289,298],[148,1790],{"className":1791,"style":308},[307],[148,1793,1795],{"className":1794},[337,199],[148,1796,1798,1931],{"className":1797},[341,342],[148,1799,1801,1928],{"className":1800},[346],[148,1802,1804,1822],{"className":1803,"style":351},[350],[148,1805,1807,1810],{"style":1806},"top:-1.3798em;",[148,1808],{"className":1809,"style":359},[358],[148,1811,1813],{"className":1812},[363,364,365,366],[148,1814,1816],{"className":1815},[289,366],[148,1817,1819],{"className":1818},[289,373,366],[148,1820,1547],{"className":1821},[289,366],[148,1823,1824,1827],{"style":380},[148,1825],{"className":1826,"style":359},[358],[148,1828,1830],{"className":1829},[337,199],[148,1831,1833,1920],{"className":1832},[341,342],[148,1834,1836,1917],{"className":1835},[346],[148,1837,1839,1869],{"className":1838,"style":351},[350],[148,1840,1842,1845],{"className":1841,"style":600},[399],[148,1843],{"className":1844,"style":359},[358],[148,1846,1848,1855,1862],{"className":1847,"style":408},[407],[148,1849,1851],{"className":1850,"style":413},[412],[415,1852,1853],{"xmlns":417,"width":418,"height":419,"viewBox":420,"preserveAspectRatio":421},[423,1854],{"d":425},[148,1856,1858],{"className":1857,"style":413},[429],[415,1859,1860],{"xmlns":417,"width":418,"height":419,"viewBox":420,"preserveAspectRatio":432},[423,1861],{"d":435},[148,1863,1865],{"className":1864,"style":413},[439],[415,1866,1867],{"xmlns":417,"width":418,"height":419,"viewBox":420,"preserveAspectRatio":442},[423,1868],{"d":445},[148,1870,1871,1874],{"style":380},[148,1872],{"className":1873,"style":359},[358],[148,1875,1877],{"className":1876},[289],[148,1878,1880,1883],{"className":1879},[289],[148,1881,175],{"className":1882},[289,290],[148,1884,1886],{"className":1885},[463],[148,1887,1889,1909],{"className":1888},[341,342],[148,1890,1892,1906],{"className":1891},[346],[148,1893,1895],{"className":1894,"style":654},[350],[148,1896,1897,1900],{"style":476},[148,1898],{"className":1899,"style":480},[358],[148,1901,1903],{"className":1902},[363,364,365,366],[148,1904,34],{"className":1905},[289,298,366],[148,1907,491],{"className":1908},[490],[148,1910,1912],{"className":1911},[346],[148,1913,1915],{"className":1914,"style":675},[350],[148,1916],{},[148,1918,491],{"className":1919},[490],[148,1921,1923],{"className":1922},[346],[148,1924,1926],{"className":1925,"style":687},[350],[148,1927],{},[148,1929,491],{"className":1930},[490],[148,1932,1934],{"className":1933},[346],[148,1935,1938],{"className":1936,"style":1937},[350],"height:1.7563em;",[148,1939],{},[148,1941],{"className":1942,"style":528},[307],[148,1944,220],{"className":1945},[532],[148,1947],{"className":1948,"style":528},[307],[148,1950,1952,1956,1959,1962],{"className":1951},[280],[148,1953],{"className":1954,"style":1955},[284],"height:2.4516em;vertical-align:-1.7016em;",[148,1957,243],{"className":1958,"style":721},[289,298],[148,1960],{"className":1961,"style":308},[307],[148,1963,1965],{"className":1964},[337,199],[148,1966,1968,2104],{"className":1967},[341,342],[148,1969,1971,2101],{"className":1970},[346],[148,1972,1974,1992],{"className":1973,"style":737},[350],[148,1975,1977,1980],{"style":1976},"top:-1.4345em;",[148,1978],{"className":1979,"style":359},[358],[148,1981,1983],{"className":1982},[363,364,365,366],[148,1984,1986],{"className":1985},[289,366],[148,1987,1989],{"className":1988},[289,373,366],[148,1990,1572],{"className":1991},[289,366],[148,1993,1994,1997],{"style":380},[148,1995],{"className":1996,"style":359},[358],[148,1998,2000],{"className":1999},[337,199],[148,2001,2003,2093],{"className":2002},[341,342],[148,2004,2006,2090],{"className":2005},[346],[148,2007,2009,2039],{"className":2008,"style":737},[350],[148,2010,2012,2015],{"className":2011,"style":776},[399],[148,2013],{"className":2014,"style":359},[358],[148,2016,2018,2025,2032],{"className":2017,"style":408},[407],[148,2019,2021],{"className":2020,"style":413},[412],[415,2022,2023],{"xmlns":417,"width":418,"height":419,"viewBox":420,"preserveAspectRatio":421},[423,2024],{"d":425},[148,2026,2028],{"className":2027,"style":413},[429],[415,2029,2030],{"xmlns":417,"width":418,"height":419,"viewBox":420,"preserveAspectRatio":432},[423,2031],{"d":435},[148,2033,2035],{"className":2034,"style":413},[439],[415,2036,2037],{"xmlns":417,"width":418,"height":419,"viewBox":420,"preserveAspectRatio":442},[423,2038],{"d":445},[148,2040,2041,2044],{"style":380},[148,2042],{"className":2043,"style":359},[358],[148,2045,2047,2050],{"className":2046},[289],[148,2048,815],{"className":2049},[289],[148,2051,2053,2056],{"className":2052},[289],[148,2054,252],{"className":2055},[289],[148,2057,2059],{"className":2058},[463],[148,2060,2062,2082],{"className":2061},[341,342],[148,2063,2065,2079],{"className":2064},[346],[148,2066,2068],{"className":2067,"style":834},[350],[148,2069,2070,2073],{"style":476},[148,2071],{"className":2072,"style":480},[358],[148,2074,2076],{"className":2075},[363,364,365,366],[148,2077,262],{"className":2078},[289,366],[148,2080,491],{"className":2081},[490],[148,2083,2085],{"className":2084},[346],[148,2086,2088],{"className":2087,"style":498},[350],[148,2089],{},[148,2091,491],{"className":2092},[490],[148,2094,2096],{"className":2095},[346],[148,2097,2099],{"className":2098,"style":866},[350],[148,2100],{},[148,2102,491],{"className":2103},[490],[148,2105,2107],{"className":2106},[346],[148,2108,2111],{"className":2109,"style":2110},[350],"height:1.7016em;",[148,2112],{},[34,2114,2115],{},[37,2116,2117],{},"How the pieces fit together",[129,2119,2120,2123,2126],{},[96,2121,2122],{},"A DNN learns the representation of the solution",[96,2124,2125],{},"Automatic differentiation computes derivatives accurately",[96,2127,2128],{},"ℓ₀ sparse regression picks out the governing terms",[34,2130,2131],{},[37,2132,2133],{},"Architecture",[129,2135,2136,2142],{},[96,2137,2138,2141],{},[74,2139,2140],{},"Single dataset",": the DNN directly models the solution function",[96,2143,2144,2147,2148],{},[74,2145,2146],{},"Multiple datasets",": a \"root-branch\" DNN handles different initial\u002Fboundary conditions\n",[129,2149,2150,2153],{},[96,2151,2152],{},"a shared root network captures common features",[96,2154,2155],{},"independent branch networks adapt to each condition",[34,2157,2158,2161,2162,2192,2193,2222],{},[37,2159,2160],{},"Optimisation (ADO)"," — alternating direction optimisation trains the network parameters ",[148,2163,2165,2179],{"className":2164},[155],[148,2166,2168],{"className":2167},[159],[161,2169,2170],{"xmlns":163},[166,2171,2172,2176],{},[169,2173,2174],{},[172,2175,183],{},[269,2177,2178],{"encoding":271},"\\theta",[148,2180,2182],{"className":2181,"ariaHidden":186},[276],[148,2183,2185,2189],{"className":2184},[280],[148,2186],{"className":2187,"style":2188},[284],"height:0.6944em;",[148,2190,183],{"className":2191,"style":299},[289,298]," and the PDE coefficients ",[148,2194,2196,2210],{"className":2195},[155],[148,2197,2199],{"className":2198},[159],[161,2200,2201],{"xmlns":163},[166,2202,2203,2207],{},[169,2204,2205],{},[172,2206,191],{"mathvariant":190},[269,2208,2209],{"encoding":271},"\\Lambda",[148,2211,2213],{"className":2212,"ariaHidden":186},[276],[148,2214,2216,2219],{"className":2215},[280],[148,2217],{"className":2218,"style":351},[284],[148,2220,191],{"className":2221},[289]," in turn, which sidesteps the NP-hard ℓ₀ problem.",[12,2224,2226],{"id":2225},"results","Results",[934,2228,2229,2245],{},[937,2230,2231],{},[940,2232,2233,2236,2239,2242],{},[943,2234,2235],{},"PDE",[943,2237,2238],{},"Noise",[943,2240,2241],{},"Coefficient error",[943,2243,2244],{},"Data used",[956,2246,2247,2258,2269,2280],{},[940,2248,2249,2252,2254,2256],{},[961,2250,2251],{},"Burgers",[961,2253,966],{},[961,2255,969],{},[961,2257,972],{},[940,2259,2260,2263,2265,2267],{},[961,2261,2262],{},"Kuramoto–Sivashinsky (chaotic)",[961,2264,966],{},[961,2266,982],{},[961,2268,985],{},[940,2270,2271,2274,2276,2278],{},[961,2272,2273],{},"Navier–Stokes",[961,2275,966],{},[961,2277,995],{},[961,2279,998],{},[940,2281,2282,2285,2287,2289],{},[961,2283,2284],{},"Schrödinger",[961,2286,966],{},[961,2288,1008],{},[961,2290,1011],{},[34,2292,2293],{},"Multiple-dataset settings were tested on Burgers with small-viscosity shock waves and on the FitzHugh–Nagumo neuron model.",[34,2295,2296],{},"The real-world test is the interesting one: on a cell migration and proliferation experiment, the method recovered the Fisher–Kolmogorov model,",[148,2298,2300],{"className":2299},[151],[148,2301,2303,2358],{"className":2302},[155],[148,2304,2306],{"className":2305},[159],[161,2307,2308],{"xmlns":163,"display":164},[166,2309,2310,2356],{},[169,2311,2312,2318,2320,2322,2332,2334,2340,2342,2344,2350],{},[204,2313,2314,2316],{},[172,2315,1050],{},[172,2317,1053],{},[177,2319,197],{},[172,2321,1058],{},[204,2323,2324,2326],{},[172,2325,1050],{},[169,2327,2328,2330],{},[172,2329,1067],{},[172,2331,1067],{},[177,2333,220],{},[204,2335,2336,2338],{},[172,2337,1076],{},[260,2339,43],{},[172,2341,1050],{},[177,2343,220],{},[204,2345,2346,2348],{},[172,2347,1076],{},[260,2349,50],{},[1090,2351,2352,2354],{},[172,2353,1050],{},[260,2355,50],{},[269,2357,1098],{"encoding":271},[148,2359,2361,2416,2477,2535],{"className":2360,"ariaHidden":186},[276],[148,2362,2364,2367,2407,2410,2413],{"className":2363},[280],[148,2365],{"className":2366,"style":1108},[284],[148,2368,2370,2373],{"className":2369},[289],[148,2371,1050],{"className":2372},[289,298],[148,2374,2376],{"className":2375},[463],[148,2377,2379,2399],{"className":2378},[341,342],[148,2380,2382,2396],{"className":2381},[346],[148,2383,2385],{"className":2384,"style":1127},[350],[148,2386,2387,2390],{"style":476},[148,2388],{"className":2389,"style":480},[358],[148,2391,2393],{"className":2392},[363,364,365,366],[148,2394,1053],{"className":2395},[289,298,366],[148,2397,491],{"className":2398},[490],[148,2400,2402],{"className":2401},[346],[148,2403,2405],{"className":2404,"style":498},[350],[148,2406],{},[148,2408],{"className":2409,"style":319},[307],[148,2411,197],{"className":2412},[323],[148,2414],{"className":2415,"style":319},[307],[148,2417,2419,2422,2425,2468,2471,2474],{"className":2418},[280],[148,2420],{"className":2421,"style":1165},[284],[148,2423,1058],{"className":2424,"style":1169},[289,298],[148,2426,2428,2431],{"className":2427},[289],[148,2429,1050],{"className":2430},[289,298],[148,2432,2434],{"className":2433},[463],[148,2435,2437,2460],{"className":2436},[341,342],[148,2438,2440,2457],{"className":2439},[346],[148,2441,2443],{"className":2442,"style":654},[350],[148,2444,2445,2448],{"style":476},[148,2446],{"className":2447,"style":480},[358],[148,2449,2451],{"className":2450},[363,364,365,366],[148,2452,2454],{"className":2453},[289,366],[148,2455,1202],{"className":2456},[289,298,366],[148,2458,491],{"className":2459},[490],[148,2461,2463],{"className":2462},[346],[148,2464,2466],{"className":2465,"style":498},[350],[148,2467],{},[148,2469],{"className":2470,"style":528},[307],[148,2472,220],{"className":2473},[532],[148,2475],{"className":2476,"style":528},[307],[148,2478,2480,2483,2523,2526,2529,2532],{"className":2479},[280],[148,2481],{"className":2482,"style":1229},[284],[148,2484,2486,2489],{"className":2485},[289],[148,2487,1076],{"className":2488},[289,298],[148,2490,2492],{"className":2491},[463],[148,2493,2495,2515],{"className":2494},[341,342],[148,2496,2498,2512],{"className":2497},[346],[148,2499,2501],{"className":2500,"style":834},[350],[148,2502,2503,2506],{"style":476},[148,2504],{"className":2505,"style":480},[358],[148,2507,2509],{"className":2508},[363,364,365,366],[148,2510,43],{"className":2511},[289,366],[148,2513,491],{"className":2514},[490],[148,2516,2518],{"className":2517},[346],[148,2519,2521],{"className":2520,"style":498},[350],[148,2522],{},[148,2524,1050],{"className":2525},[289,298],[148,2527],{"className":2528,"style":528},[307],[148,2530,220],{"className":2531},[532],[148,2533],{"className":2534,"style":528},[307],[148,2536,2538,2541,2581],{"className":2537},[280],[148,2539],{"className":2540,"style":1288},[284],[148,2542,2544,2547],{"className":2543},[289],[148,2545,1076],{"className":2546},[289,298],[148,2548,2550],{"className":2549},[463],[148,2551,2553,2573],{"className":2552},[341,342],[148,2554,2556,2570],{"className":2555},[346],[148,2557,2559],{"className":2558,"style":834},[350],[148,2560,2561,2564],{"style":476},[148,2562],{"className":2563,"style":480},[358],[148,2565,2567],{"className":2566},[363,364,365,366],[148,2568,50],{"className":2569},[289,366],[148,2571,491],{"className":2572},[490],[148,2574,2576],{"className":2575},[346],[148,2577,2579],{"className":2578,"style":498},[350],[148,2580],{},[148,2582,2584,2587],{"className":2583},[289],[148,2585,1050],{"className":2586},[289,298],[148,2588,2590],{"className":2589},[463],[148,2591,2593],{"className":2592},[341],[148,2594,2596],{"className":2595},[346],[148,2597,2599],{"className":2598,"style":1347},[350],[148,2600,2601,2604],{"style":1350},[148,2602],{"className":2603,"style":480},[358],[148,2605,2607],{"className":2606},[363,364,365,366],[148,2608,50],{"className":2609},[289,366],[12,2611,2613],{"id":2612},"why-it-matters","Why it matters",[34,2615,2616,2619],{},[37,2617,2618],{},"It breaks the data wall."," Accurate discovery under 10% noise with less than 1% of the original grid is the headline result.",[34,2621,2622,2625],{},[37,2623,2624],{},"It stays interpretable."," The DNN provides a global approximation, while the sparse PDE constrains the network.",[34,2627,2628,2631],{},[37,2629,2630],{},"It generalises to expensive domains"," — fluid mechanics, biological migration, climate prediction — where you can never collect a dense dataset.",[12,2633,2635],{"id":2634},"my-take","My take",[34,2637,2638,2639,2642],{},"The valuable idea here is not \"use a neural network\". It is the ",[37,2640,2641],{},"separation of concerns",": let the network own the function, let sparse regression own the equation, and let AD bridge the two. Each part is something you can reason about and check.",[34,2644,2645],{},"The weakness is the same one every physics-informed method shares: the library of candidate terms is a human decision. If the true governing term is not in the candidate set, no amount of optimisation will find it.",{"title":1380,"searchDepth":1381,"depth":1381,"links":2647},[2648,2649,2650,2651,2652,2653],{"id":1439,"depth":1381,"text":1440},{"id":1470,"depth":1381,"text":1471},{"id":1480,"depth":1381,"text":1481},{"id":2225,"depth":1381,"text":2226},{"id":2612,"depth":1381,"text":2613},{"id":2634,"depth":1381,"text":2635},"PiDL recovers the PDEs behind nonlinear spatio-temporal systems from sparse, noisy observations. Notes on its architecture, training strategy, and results on Burgers and Navier–Stokes.","en",{},"\u002Fblog\u002Fen\u002Farxiv-2005-03448v3",{"title":1416,"description":2654},"blog\u002Fen\u002Farxiv-2005-03448v3",[1409,1410,1411],"RIokw_sIVtYhgrTpZy8_PrCcQPPaMadK3X9n8I9ivck",{"id":2663,"company":2664,"extension":2675,"meta":2676,"services":2677,"social":2755,"stem":2756,"__hash__":2757},"site\u002Fsite.yml",{"name":2665,"nameEn":2666,"email":2667,"address":2668,"addressEn":2669,"lineUrl":2670,"lineId":2671,"duns":2672,"youtube":2673,"youtubeId":2674},"銡策數位有限公司","Jecision Digital Co., Ltd.","contact@jecision.com","231 新北市新店區北新路二段 262 號 3 樓之 3","3F.-3, No. 262, Sec. 2, Beixin Rd., Xindian Dist., New Taipei City 231, Taiwan","https:\u002F\u002Fline.me\u002FR\u002Fti\u002Fp\u002F@374qnumi","@374qnumi","963905705","https:\u002F\u002Fwww.youtube.com\u002Fshorts\u002F6UE0WTmyjsg","6UE0WTmyjsg","yml",{},[2678,2702,2728],{"key":2679,"zh":2680,"en":2691},"audit",{"title":2681,"tagline":2682,"summary":2683,"points":2684,"deliverable":2688,"detail":2689,"duration":2690},"企業 AI 導入健檢","先看清楚，再花錢","60 分鐘線上訪談，盤點最耗時、最依賴人力的營運環節；我們會告訴你哪些工作現在就能交給 AI、哪些還不成熟。",[2685,2686,2687],"現況盤點：梳理日常營運中最耗時、最依賴人力的環節","痛點診斷：判斷哪些工作現在就能交給 AI，哪些建議先保留人力","可行性評估：依你的業務特性與預算，篩選適合的模型與工具","《企業 AI 導入建議報告書》（PDF）","含 1–2 個「開發成本最低、成效最明顯」的優先項目、技術棧建議、預期節省的人力時間與下一步","訪談 60 分鐘，報告書 3 個工作天內交付",{"title":2692,"tagline":2693,"summary":2694,"points":2695,"deliverable":2699,"detail":2700,"duration":2701},"AI Readiness Audit","Look before you spend","A 60-minute remote audit of the workflows that eat the most time and people-power — and an honest answer on what AI can take over today, and what it cannot.",[2696,2697,2698],"Audit the daily workflows that consume the most time and headcount","Triage which tasks AI can own today, and which should stay human","Shortlist the models and tools that fit your budget and constraints","AI Adoption Recommendation Report (PDF)","One or two highest-ROI entry points, a recommended tech stack, expected hours saved, and the concrete next step","60-minute call, report delivered within 3 business days",{"key":2703,"zh":2704,"en":2715},"prompt",{"title":2705,"tagline":2706,"summary":2707,"points":2708,"deliverable":2712,"detail":2713,"duration":2714},"提示詞工程","讓 AI 穩定產出，而不是碰運氣","針對你的實際任務設計提示詞與多步驟工作流，並做極限測試，確保換了資料、換了人操作也不會走鐘。",[2709,2710,2711],"客製提示詞：ChatGPT、Claude、Gemini、Midjourney 等主流工具的專屬模板","除錯與優化：修掉幻覺、格式跑掉、時好時壞的問題","工作流建置：把複雜任務拆成多步驟提示詞與可重複的 SOP","提示詞文件或設定檔","附使用說明與範例，並提供 1 對 1 教學，確保你與團隊能自己維護","依任務複雜度報價，通常 3–7 個工作天",{"title":2716,"tagline":2717,"summary":2718,"points":2719,"deliverable":2725,"detail":2726,"duration":2727},"Prompt Engineering","Make AI produce reliably, not by luck","Prompt sets and multi-step workflows designed for your actual tasks, stress-tested so they keep working when the data or the operator changes.",[2720,2721,2723],"Custom prompts for ChatGPT, Claude, Gemini, Midjourney and friends",{"Debugging and tuning":2722},"kill hallucinations, broken formats and flaky output",{"Workflow design":2724},"break complex work into repeatable multi-step SOPs","Prompt library or exportable config","Ships with usage notes and examples, plus a 1:1 walkthrough so your team can maintain it","Quoted by complexity, typically 3–7 business days",{"key":2729,"zh":2730,"en":2741},"poc",{"title":2731,"tagline":2732,"summary":2733,"points":2734,"deliverable":2738,"detail":2739,"duration":2740},"極速 PoC","五天，看到能跑的東西","不談架構藍圖，直接做一個能動的雛形。五個工作天內交付線上可操作的智能體，用你真實的資料驗證。",[2735,2736,2737],"Day 1：15 分鐘對焦，鎖定單一痛點與資料盤點","Day 2–4：模型串接、邏輯設計與雛形開發","Day 5：Demo 展示與 AI 擴展建議書","可線上操作的智能體雛形","交付範圍為概念驗證與展示，不含後續維運或大規模伺服器部署；要進階成完整系統會另行報價","5 個工作天",{"title":2742,"tagline":2743,"summary":2744,"points":2745,"deliverable":2752,"detail":2753,"duration":2754},"Rapid PoC","Five days to something that runs","No architecture decks — a working prototype built in five business days, validated against your real data.",[2746,2748,2750],{"Day 1":2747},"15-minute scoping call, one pain point, data inventory",{"Day 2–4":2749},"model integration, logic design, prototype build",{"Day 5":2751},"live demo plus an AI expansion recommendation","A runnable online agent prototype","Proof of concept and demo scope only; ongoing operations and large-scale deployment are quoted separately","5 business days",[],"site","XTwARfVAOcrXNiKehsM6V6cdkd2HXy4d5yZVrSWrI3Q",[2759,3024],{"id":2760,"title":2761,"author":7,"body":2762,"cover":3011,"coverAlt":3012,"date":3013,"description":3014,"draft":1399,"extension":1400,"featured":1399,"lang":2655,"meta":3015,"navigation":1403,"path":3016,"seo":3017,"slug":3018,"stem":3019,"tags":3020,"translationKey":3018,"updated":1412,"__hash__":3023},"blog\u002Fblog\u002Fen\u002Farxiv-2510-27688v1.md","CALM, or why token-by-token generation hit a wall",{"type":9,"value":2763,"toc":3003},[2764,2777,2781,2784,2787,2801,2804,2808,2815,2822,2871,2876,2880,2886,2892,2897,2908,2911,2916,2930,2934,2951,2953,2974,2976,2983,2986,3000],[34,2765,2766,2769,2770,1424,2772],{},[37,2767,2768],{},"From"," — Tencent WeChat AI × Tsinghua University\n",[37,2771,1423],{},[17,2773,2776],{"href":2774,"rel":2775},"https:\u002F\u002Farxiv.org\u002Fhtml\u002F2510.27688v1",[21],"arXiv:2510.27688",[12,2778,2780],{"id":2779},"the-wall","The wall",[34,2782,2783],{},"Every large language model today predicts one token at a time. Each token costs a full forward pass, so cost grows linearly — and with attention, worse — as sequences get longer.",[34,2785,2786],{},"Text representation has already moved from characters to subwords (BPE), but the discrete representation itself is running out of room:",[129,2788,2789,2795],{},[96,2790,2791,2794],{},[37,2792,2793],{},"Vocabulary inflation",": to cover more meaning, vocabularies grew from 32k to 256k. The final layer's compute and memory cost exploded with them.",[96,2796,2797,2800],{},[37,2798,2799],{},"Low information density",": each discrete token carries roughly 15–18 bits. The model has enormous compute and spends it on the labour-intensive job of predicting one low-information token at a time.",[34,2802,2803],{},"CALM's premise: stop picking from a vocabulary, and predict a point in continuous space instead.",[12,2805,2807],{"id":2806},"the-idea","The idea",[34,2809,2810,2811,2814],{},"An autoencoder compresses ",[37,2812,2813],{},"K discrete tokens into one high-dimensional continuous vector",", so sequence length drops from T to T\u002FK and the number of generation steps drops by the same factor.",[34,2816,2817,2818,2821],{},"It trades more ",[74,2819,2820],{},"semantic bandwidth"," per step for fewer steps.",[934,2823,2824,2836],{},[937,2825,2826],{},[940,2827,2828,2830,2833],{},[943,2829],{},[943,2831,2832],{},"Discrete",[943,2834,2835],{},"CALM (continuous)",[956,2837,2838,2849,2860],{},[940,2839,2840,2843,2846],{},[961,2841,2842],{},"Scaling",[961,2844,2845],{},"vocabulary grows exponentially",[961,2847,2848],{},"vector dimension grows smoothly",[940,2850,2851,2854,2857],{},[961,2852,2853],{},"Objective",[961,2855,2856],{},"pick one of hundreds of thousands",[961,2858,2859],{},"predict coordinates in a continuous space",[940,2861,2862,2865,2868],{},[961,2863,2864],{},"Sequence",[961,2866,2867],{},"long T",[961,2869,2870],{},"short T\u002FK",[69,2872,2873],{},[34,2874,2875],{},"A useful mental image: conventional generation is searching a warehouse with hundreds of thousands of drawers. CALM draws coordinates.",[12,2877,2879],{"id":2878},"four-pillars","Four pillars",[34,2881,2882,2885],{},[37,2883,2884],{},"1. A high-fidelity autoencoder."," Compresses K tokens into 128+ dimensions at roughly 75M parameters of overhead, with reconstruction accuracy above 99.9%. Without this, everything downstream is garbage.",[34,2887,2888,2891],{},[37,2889,2890],{},"2. An energy-based head, no likelihood."," Instead of softmax over a vocabulary, the model regresses the next target vector directly — a single step, not the dozens or hundreds of denoising iterations a diffusion model needs.",[34,2893,2894],{},[37,2895,2896],{},"3. Three tricks for robustness in continuous space.",[129,2898,2899,2902,2905],{},[96,2900,2901],{},"Variational regularisation, to smooth folds and cliffs in the latent space",[96,2903,2904],{},"KL clipping, to stop dimensions collapsing",[96,2906,2907],{},"Double dropout, for extra noise tolerance",[34,2909,2910],{},"Together they build a smooth, elastic landscape where a small prediction error does not turn into gibberish.",[34,2912,2913],{},[37,2914,2915],{},"4. Evaluation without probabilities.",[129,2917,2918,2924],{},[96,2919,2920,2923],{},[37,2921,2922],{},"BrierLM",": perplexity does not apply when there is no likelihood. CALM uses a Brier score (accuracy minus collision rate) as its yardstick, which correlates strongly and negatively with cross-entropy loss.",[96,2925,2926,2929],{},[37,2927,2928],{},"Black-box temperature",": with no probability denominator to tune, batch size N becomes the temperature knob. Small N is cold and deterministic, large N is hot and diverse. With rejection sampling, it behaves much like conventional temperature.",[12,2931,2933],{"id":2932},"the-loop","The loop",[93,2935,2936,2939,2942,2945,2948],{},[96,2937,2938],{},"Input: discrete tokens",[96,2940,2941],{},"Compress: the encoder turns every K tokens into one vector",[96,2943,2944],{},"Process: the transformer runs attention over a much shorter sequence",[96,2946,2947],{},"Predict: the head emits the next continuous vector in one step",[96,2949,2950],{},"Restore: the decoder expands it back into K tokens",[12,2952,2226],{"id":2225},[129,2954,2955,2958,2964,2971],{},[96,2956,2957],{},"With K=4: training compute down 44%, inference compute down 34%, generation steps from T to T\u002F4",[96,2959,2960,2963],{},[37,2961,2962],{},"Larger but cheaper",": 371M parameters (slightly larger than the baseline) with equal or better quality and less total compute",[96,2965,2966,2967,2970],{},"A new scaling axis: ",[37,2968,2969],{},"information density",". As semantic bandwidth grows, CALM reaches a higher Brier score at lower compute — pushing past the old compute–quality frontier",[96,2972,2973],{},"An unusual training curve: performance climbs sharply after ~200k steps and then overtakes conventional models. Learning continuous structure seems to need a long runway before it pays off",[12,2975,2635],{"id":2634},[34,2977,2978,2979,2982],{},"The framing is the contribution: ",[37,2980,2981],{},"the vocabulary was never the point",". Discrete tokens are a compression artifact of text, and once you accept that, predicting a continuous vector is the obvious next move.",[34,2984,2985],{},"Two things to watch before believing it replaces anything:",[93,2987,2988,2994],{},[96,2989,2990,2993],{},[37,2991,2992],{},"Reconstruction fidelity is a hard ceiling."," 99.9% sounds great until you multiply it across a long document.",[96,2995,2996,2999],{},[37,2997,2998],{},"Sampling control moves to a different knob."," Batch-size-as-temperature works, but it means your serving stack has to be redesigned around it.",[34,3001,3002],{},"For deployment on constrained hardware, a 4× reduction in generation steps is the most interesting number here — that is the kind of win that changes what fits on a device.",{"title":1380,"searchDepth":1381,"depth":1381,"links":3004},[3005,3006,3007,3008,3009,3010],{"id":2779,"depth":1381,"text":2780},{"id":2806,"depth":1381,"text":2807},{"id":2878,"depth":1381,"text":2879},{"id":2932,"depth":1381,"text":2933},{"id":2225,"depth":1381,"text":2226},{"id":2634,"depth":1381,"text":2635},"\u002Fimages\u002Fblog\u002F2510.27688v1.webp","Continuous autoregressive language models (CALM)","2026-08-06","CALM compresses K discrete tokens into one continuous vector, cutting generation steps to 1\u002FK. Notes on the autoencoder, the likelihood-free head, and what it actually buys you.",{},"\u002Fblog\u002Fen\u002Farxiv-2510-27688v1",{"title":2761,"description":3014},"arxiv-2510-27688v1","blog\u002Fen\u002Farxiv-2510-27688v1",[1409,3021,3022],"llm","efficiency","6E96cFhns3BbhpqUak9RrqhAZpia6m0Go_yTYpgqN2g",{"id":3025,"title":3026,"author":7,"body":3027,"cover":3354,"coverAlt":3355,"date":3356,"description":3357,"draft":1399,"extension":1400,"featured":1399,"lang":2655,"meta":3358,"navigation":1403,"path":3359,"seo":3360,"slug":3361,"stem":3362,"tags":3363,"translationKey":3361,"updated":1412,"__hash__":3366},"blog\u002Fblog\u002Fen\u002Farxiv-2506-13777v1.md","Physics-informed AI for complex urban systems",{"type":9,"value":3028,"toc":3347},[3029,3041,3045,3048,3051,3058,3062,3176,3183,3187,3281,3284,3295,3299,3302,3322,3325,3329,3332,3338,3344],[34,3030,3031,1424,3033,3038,3040],{},[37,3032,1423],{},[17,3034,3037],{"href":3035,"rel":3036},"https:\u002F\u002Farxiv.org\u002Fhtml\u002F2506.13777v1",[21],"arXiv:2506.13777",[37,3039,1431],{}," — A Survey of Physics-Informed AI for Complex Urban Systems",[12,3042,3044],{"id":3043},"why-this-survey-exists","Why this survey exists",[34,3046,3047],{},"Cities are the awkward middle ground of modelling. Pure physics cannot capture human behaviour; pure data-driven models cannot explain themselves, and they break exactly when conditions change — which is when a city most needs a prediction.",[34,3049,3050],{},"Physics-informed AI (PIAI) is the attempt to have both. Its selling point is a model that respects conservation laws and heat transfer while still learning from data.",[34,3052,3053,3054,3057],{},"The survey's contribution is organisational: ",[37,3055,3056],{},"a taxonomy of three paradigms and seven methods",", a review of applications across eight urban subsystems, and an honest list of what is still missing.",[12,3059,3061],{"id":3060},"the-taxonomy","The taxonomy",[934,3063,3064,3080],{},[937,3065,3066],{},[940,3067,3068,3071,3074,3077],{},[943,3069,3070],{},"Paradigm",[943,3072,3073],{},"Method",[943,3075,3076],{},"Core idea",[943,3078,3079],{},"Typical techniques",[956,3081,3082,3096,3109,3122,3136,3149,3163],{},[940,3083,3084,3087,3090,3093],{},[961,3085,3086],{},"Physics-integrated AI",[961,3088,3089],{},"1. PINN with loss function",[961,3091,3092],{},"physics constraints as a loss term",[961,3094,3095],{},"PINN, CPINN, PCNN",[940,3097,3098,3100,3103,3106],{},[961,3099],{},[961,3101,3102],{},"2. PINN with weight initialisation",[961,3104,3105],{},"initialise weights from physical knowledge",[961,3107,3108],{},"transfer learning, self-supervised pretraining",[940,3110,3111,3113,3116,3119],{},[961,3112],{},[961,3114,3115],{},"3. PINN with architecture design",[961,3117,3118],{},"bake physical structure into the network",[961,3120,3121],{},"symmetry encoding, Hamiltonian structure",[940,3123,3124,3127,3130,3133],{},[961,3125,3126],{},"Hybrid",[961,3128,3129],{},"4. Sequential ensemble",[961,3131,3132],{},"physics and AI cooperate in stages",[961,3134,3135],{},"residual modelling, pipeline architectures",[940,3137,3138,3140,3143,3146],{},[961,3139],{},[961,3141,3142],{},"5. Parallel ensemble",[961,3144,3145],{},"run both, fuse the outputs",[961,3147,3148],{},"dynamic weighting, model combination",[940,3150,3151,3154,3157,3160],{},[961,3152,3153],{},"AI-integrated physics",[961,3155,3156],{},"6. AI-discovered physics",[961,3158,3159],{},"let AI find the governing law",[961,3161,3162],{},"SINDy, symbolic regression",[940,3164,3165,3167,3170,3173],{},[961,3166],{},[961,3168,3169],{},"7. Neural physics model",[961,3171,3172],{},"replace the hard-to-model part with a network",[961,3174,3175],{},"neural closures for turbulence",[34,3177,3178,3179,3182],{},"The useful distinction is ",[37,3180,3181],{},"where physics enters",": as a penalty, as an initialisation, as architecture, as a co-model, or not at all (physics emerges from data).",[12,3184,3186],{"id":3185},"where-it-is-used","Where it is used",[934,3188,3189,3202],{},[937,3190,3191],{},[940,3192,3193,3196,3199],{},[943,3194,3195],{},"Subsystem",[943,3197,3198],{},"Physical law \u002F model",[943,3200,3201],{},"Example application",[956,3203,3204,3215,3226,3237,3248,3259,3270],{},[940,3205,3206,3209,3212],{},[961,3207,3208],{},"Energy",[961,3210,3211],{},"power flow equations, pressure balance",[961,3213,3214],{},"grid optimisation, renewable integration",[940,3216,3217,3220,3223],{},[961,3218,3219],{},"Environment",[961,3221,3222],{},"Gaussian plume, SWMM hydrology",[961,3224,3225],{},"air quality, flood simulation",[940,3227,3228,3231,3234],{},[961,3229,3230],{},"Economy",[961,3232,3233],{},"network flow, Huff choice model",[961,3235,3236],{},"supply chain, demand forecasting",[940,3238,3239,3242,3245],{},[961,3240,3241],{},"Transport",[961,3243,3244],{},"LWR traffic flow, car-following",[961,3246,3247],{},"state estimation, signal control",[940,3249,3250,3253,3256],{},[961,3251,3252],{},"Information",[961,3254,3255],{},"Shannon capacity, ray tracing",[961,3257,3258],{},"information diffusion in social networks",[940,3260,3261,3264,3267],{},[961,3262,3263],{},"Public services",[961,3265,3266],{},"SIR epidemic, social force",[961,3268,3269],{},"outbreak forecasting, crowd evacuation",[940,3271,3272,3275,3278],{},[961,3273,3274],{},"Emergency",[961,3276,3277],{},"social force, spatio-temporal decay",[961,3279,3280],{},"disaster dynamics, post-disaster mobility",[34,3282,3283],{},"Highlights from the review: PINNs reconstructing wind farms and predicting solar output; Kirchhoff-constrained networks accelerating optimal power flow; diffusion–advection equations fused with neural networks for pollutant forecasting; LWR\u002FARZ models combined with deep learning to estimate traffic state from sparse sensors; neural ODEs modelling opinion dynamics; epidemiological neural networks fusing multi-source data for outbreak prediction.",[34,3285,3286,3287,3290,3291,3294],{},"The last section is the most interesting: ",[37,3288,3289],{},"cross-domain coupling"," — graph neural networks linking flood dynamics with transport models to simulate a city under extreme weather — and ",[37,3292,3293],{},"multi-scale integration",", from building-level thermal models up to regional climate.",[12,3296,3298],{"id":3297},"the-open-problems","The open problems",[34,3300,3301],{},"The survey is refreshingly blunt about what is unsolved:",[129,3303,3304,3310,3316],{},[96,3305,3306,3309],{},[37,3307,3308],{},"Weak theory"," — convergence and generalisation of PINNs lack rigorous proofs",[96,3311,3312,3315],{},[37,3313,3314],{},"Compute"," — training cost is high, which rules out real-time deployment",[96,3317,3318,3321],{},[37,3319,3320],{},"Knowledge representation"," — causal and rule-based (non-numeric) knowledge is not integrated",[34,3323,3324],{},"Proposed directions include multi-resolution architectures, active and transfer learning, combining PIAI with LLMs, and building urban digital twins or general world models.",[12,3326,3328],{"id":3327},"what-i-take-from-it","What I take from it",[34,3330,3331],{},"Two observations in the paper are worth more than the taxonomy itself:",[34,3333,3334,3337],{},[37,3335,3336],{},"Cities are the natural testbed for PIAI"," precisely because they combine physical regularity with human unpredictability. Both extremes exist in the same system, so a method that only works for one of them will fail visibly.",[34,3339,3340,3343],{},[37,3341,3342],{},"Data-rich tasks drift toward AI-led methods; decision-support tasks need physical interpretability."," That is a practical rule for choosing a method — not based on what is fashionable, but on who has to defend the output.",[34,3345,3346],{},"For anyone building city-scale models, the \"how do I choose\" section is the one to read first: physical knowledge availability, data density, and whether the deliverable is a number or a decision.",{"title":1380,"searchDepth":1381,"depth":1381,"links":3348},[3349,3350,3351,3352,3353],{"id":3043,"depth":1381,"text":3044},{"id":3060,"depth":1381,"text":3061},{"id":3185,"depth":1381,"text":3186},{"id":3297,"depth":1381,"text":3298},{"id":3327,"depth":1381,"text":3328},"\u002Fimages\u002Fblog\u002F2506.13777.1.webp","A survey of physics-informed AI for complex urban systems","2025-09-16","How do you put physics into city-scale models? This survey proposes a taxonomy of three paradigms and seven methods, reviews eight urban subsystems, and names what is still missing.",{},"\u002Fblog\u002Fen\u002Farxiv-2506-13777v1",{"title":3026,"description":3357},"arxiv-2506-13777v1","blog\u002Fen\u002Farxiv-2506-13777v1",[1409,3364,3365],"urban","survey","1gwEeKIu-vP0ABuoHpyY9xShz03tEALrPNkeNBiXGXc",1792775653599]