自己紹介：機械学習・機械発見とデータ中心的自然科学

戣䊛♧㷕 [email protected] 椚⻉㷕灇瑔䨽ꬠ倜濼腉窟さ灇瑔إٝة٦!❨ꢻ㣽"53 ⻌嵲麣㣐㷕⻉㷕⿾䘔ⶼ䧭灇瑔䬿挿 *$3F%% 荈䊹稱➜ 堣唒㷕统٥堣唒涪鋅הر٦ة⚥䗰涸荈搫猰㷕 2021䎃12剢22傈

戣䊛♧㷕 ׋ָֹ׻ ְ׍ָֻ 堣唒㷕统׾灇瑔׃גְ׷䪮遭㾊رأ ˖ ֲו׿源넝匟䋐欰ת׸ ˖ ։⻌嵲麣㣐䊨㷕灇瑔猰
⽆㡦䊨㷕 ⸋寸㹀⥋〾彁ⴓꨄך鍑ך椚锷ⴓ匿 ˖ ։❨鿪㣐⻉㷕灇瑔䨽讒㷕灇瑔猰 غ؎ؔ؎ٝؿؓوذ؍ؙأإٝة٦⸔侄 ˖ ։⻌嵲麣㣐䞔㜠猰㷕灇瑔猰㣐鋉垷濼陎Ⳣ椚灇瑔㹓Ⲥ侄䱇 ։+45ָֹֽׁ 勞俱؎ٝؿؓوذ؍ؙأ ˖ ։⻌嵲麣㣐㷕⻉㷕⿾䘔ⶼ䧭灇瑔䬿挿 *$3F%% ։椚⻉㷕灇瑔䨽ꬠ倜濼腉窟さ灇瑔إٝة٦ "*1 https://itakigawa.github.io 兛媮כ❨㣐J14稢脄灇הך鸬䵿ٓن!❨ꢻ㣽"53ח⹶⹡ J14稢脄鸬䵿⼔㷕涸ٔأؙ㔐鼘ث٦ي

戣䊛♧㷕 ׋ָֹ׻ ְ׍ָֻ הず儗ח堣唒㷕统ךِ٦ؠד׮֮׶תׅ ˖ ֲו׿源넝匟䋐欰ת׸ ˖ ։⻌嵲麣㣐䊨㷕灇瑔猰
⽆㡦䊨㷕 ⸋寸㹀⥋〾彁ⴓꨄך鍑ך椚锷ⴓ匿 ˖ ։❨鿪㣐⻉㷕灇瑔䨽讒㷕灇瑔猰 غ؎ؔ؎ٝؿؓوذ؍ؙأإٝة٦⸔侄 ˖ ։⻌嵲麣㣐䞔㜠猰㷕灇瑔猰㣐鋉垷濼陎Ⳣ椚灇瑔㹓Ⲥ侄䱇 ։+45ָֹֽׁ 勞俱؎ٝؿؓوذ؍ؙأ ˖ ։⻌嵲麣㣐㷕⻉㷕⿾䘔ⶼ䧭灇瑔䬿挿 *$3F%% ։椚⻉㷕灇瑔䨽ꬠ倜濼腉窟さ灇瑔إٝة٦ "*1 兛媮כ❨㣐J14稢脄灇הך鸬䵿ٓن!❨ꢻ㣽"53ח⹶⹡ J14稢脄鸬䵿⼔㷕涸ٔأؙ㔐鼘ث٦ي https://itakigawa.github.io

䪮遭灇瑔♳ךꟼ䗰ꨄ侔圓鸡׾⠵ֲ堣唒㷕统 ꨄ侔圓鸡 ꧊さה锷椚ծꟼ⤘ծ穈さׇծ禸⴨ծ加ծؚٓؿծ➿侧禸ծ鎉铂ծFUD 㼎韋חꨄ侔圓鸡 ˖ ⴓ㶨圓鸡 %։% ˖ ؜ظيؒؾ؜ظي
˖ 祩ꓲ ˖ لفثسةٝػؙ颵 ˖ ⾱㶨ךワ劍ꂁ縧 ˖ 锷椚װㄏ겗 ˖ ،؎ذيإحز ˖ 鎉铂ذؗأز ꟼ⤘חꨄ侔圓鸡 ˖ ⻉㷕⿾䘔穗騟طحزٙ٦ؙ ➿闐穗騟,&((זו ˖ 鼋⠗㶨طحزٙ٦ؙ ˖ ⴓ㶨湱✼⡲欽طحزٙ٦ؙ 11*ծ%5*ծ%%*ծFUD ˖ 㔓㶨穈さׇ ˖ 锷椚ⵖ秈 ˖ 㢌侧꟦ך❛✼⡲欽 ٌرٕחꨄ侔圓鸡 ˖ 寸㹀加،ٝ؟ٝـٕ 加װ啾װ%"( ˖ 鎘皾ؚٓؿ٥荈⹛䗍ⴓ禸帾㾴㷕统 ˖ ꥡ㾴圓鸡 ˖ ⴓ鍑圓鸡 ˖ 圓鸡ⵖ秈 ˖ ⵖ秈ػة٦ٝ⴨䮙堣唒㷕统鸬竲涸ז然桦窟鎘

寸㹀加،ٝ؟ٝـٕה荈⹛䗍ⴓؿٖ٦يٙ٦ؙ 帾㾴㷕统窟鎘侧椚灇瑔䨽ٔ٦ر؍ؚٝ%"5 %BUB"OBMZUJDT5BMFOUT 闌䏟 -#堣唒㷕统הر٦ة؟؎ؒٝأך植➿涸䩛岀ず䘊爡㣐㷕չ剑黝⻉岀պ痥㔐 ؜أزز٦ؙ https://www.slideshare.net/itakigawa/ss-241690171

㷕⡘׾《׏׋灇瑔ذ٦و ⳻꧊さח״׷鿇ⴓؙٓأ鄃銻ח״׷堣唒㷕统 Logical analysis of data (LAD) by Peter L.
Hammer Subclass method by Mineichi Kudo Decision Trees / CART by BFOS ̔չ挿꧊さ9ַ׵♧䠐ח㹀ת׷⳻꧊さ4 9 ָ➭ךؙٓأך 挿׾ろתזְ״ֲז噰㣐ז鿇ⴓ꧊さ9ך⴨䮙պה׃ג鍑ֻ ⸋寸㹀⥋〾彁ⴓꨄך鍑ך䚍颵ך椚锷ⴓ匿ה⸬桦涸実鍑岀剑㼭-ظٕي鍑実鍑ָ穈さׇ㉏겗 ˟㹋ꥷכ'PVSJFS խسً؎ٝד遤ֲ .*9

❨㣐⻉㷕灇瑔䨽غ؎ؔ؎ٝؿؓوذ؍ؙأإٝة٦ 㹧ꅿ䝎⯓欰 꼛鋅㝩⯓欰ך ⽆㡦锷俑䭷㼪侄㆞ ꆃ⛉㻛⯓欰 꼛鋅㝩䬪⯓欰

猰灇顤暴㹀걄㚖걄㚖殢〾殛獥涪鋅猰㷕㹧ꅿ䝎⯓欰剣䊛眍㣗⯓欰

堣唒㷕统ַ׵堣唒涪鋅פ https://doi.org/10.11517/jjsai.11.6_865 剣䊛眍㣗堣唒㷕统ַ׵堣唒涪鋅פ➂䊨濼腉㷕⠓钞䋆〾

؜ظي鍑匿ח⠵ֲ㣐ꆀ濼陎䞔㜠Ⳣ椚ך灇瑔

堣唒㷕统ַ׵堣唒涪鋅פ https://doi.org/10.11517/jjsai.11.6_865 剣䊛眍㣗堣唒㷕统ַ׵堣唒涪鋅פ➂䊨濼腉㷕⠓钞䋆〾

걄㚖ך䧭卓⳿晛暟涪鋅猰㷕הر٦ةو؎صؚٝ 䎷畍խ傈劤ך涪鋅猰㷕فٗآؙؑز剣䊛眍㣗 痥*鿇խ䱿锷ח״׷濼陎涪鋅 խխ痥畍խչ涪鋅պך猰㷕ㆸ㷕ˌ娖〷涸稆䲽ꅿ㹺㉔♧ խխ痥畍խ窟鎘涸鎸〾Ⳣ椚鎉铂13*4.✡靼歋ꥐ٥⡟谏岱➜ խխ痥畍խ✮庠ٌرַٕ׵ךٕ٦ٕ䬄⳿ˌ侧䒭ַ׵鎉铂פ剢劤խ峔٥啾歊⼪窩 խխ痥畍խ䌓秛锷椚فؚٗٓىؚٝה鏾僇酡㸣㿊劤畍⽆٥剣募⽆秀٥䎂歊罭♧ խխ痥畍խ,FZ(SBQIˌؗ٦ٙ٦س䬄⳿خ٦ַٕ׵涪鋅خ٦ٕפך㾜Ꟛ瀧㿊床٥ 㣐愮䎊欰٥靼ⰻ歊姻䕕
խխ痥畍խ*OUFMMJHFOU1BEךさ䧭הⱄⵃ欽ˌ䌓秛䱿锷ך甧㜥ַ׵⾱〡铆٥䎂歊幄 痥**鿇խ鎘皾㷕统椚锷ח㛇בֻ濼陎涪鋅 խխ痥畍խ腉⹛㷕统ה涪鋅猰㷕㸜⦓湫埠٥꼛鋅㝩խ䬪 խխ痥畍խֻ׶鵤׃؜٦يה׃גך㷕统،ٕ؞ٔؤي⚵䀤խ畍٥戣劤薉✳ խխ痥畍խ؝ٝؾُ٦ة؟؎ؒٝأך׋׭ך⽃秪ַא⸬桦涸ז؟ٝفؚٔٝ䪮岀床鴟խ屚 խխ痥畍խ㷕统،ٕ؞ٔؤيך鐰⣣♳⾱齥僻 խխ痥畍խ䎗⡦ؙٓأةؚٔٝך䞔㜠鎘皾䎗⡦圓鸡➙❁խ嵞 խխ痥畍խ4VQQPSU7FDUPS.BDIJOFח״׷ⴓ겲넝갭幄㸩

걄㚖ך䧭卓⳿晛暟涪鋅猰㷕הر٦ةو؎صؚٝ 痥***鿇խ堣唒㷕统הر٦ةو؎صؚٝח㛇בֻ濼陎涪鋅 խխ痥畍խ؝ٝؾُ٦ة佄䴂ח״׷猰㷕涸濼陎ך涪鋅1BU-BOHMFZ衼㹧ꅿխ䝎٥⚵㿊խ⥜鏬 խխ痥畍խ㷕统ַծو؎صַؚٝծٌرַؚٔٝˌ〢欰䡾㷕ַ׵ך✲⢽灇瑔 )FJLLJ.BOOJMBFUBM衼啾♴溪♧鏬 խխ痥畍խⴓ協ꣲ㹀岀׾欽ְ׋⚛⴨ؚٓؿ䱱稊ח״׷剑黝穠さٕ٦ٕך涪鋅⚥靼僇䒽٥啾♴溪♧ խխ痥畍խ濼陎涪鋅ה荈䊹穈籼㘗ך窟鎘ٌرٕ⻌䊛彁㔊龤٥垌〡濼⛒ խխ痥畍խ곁㹏ך飑顠㾶娖ַ׵ךر٦ةو؎صؚٝ濶歊⹧⤥٥⸇谏湫埠٥纷㹓遤⥋ խխ痥畍խ涪鋅ءأذيהؼُ٦وٝؒؗأػ٦زך؎ٝذؚٖ٦ءّٝ⚵㿊խ⥜٥㹧ꅿխ䝎
痥*7鿇խ㣐鋉垷侧⦼ر٦ةַ׵ך濼陎涪鋅 խխ痥畍խ㣕ꤿ㖑椔禸暟椚㷕פך濼陎涪鋅ך䘔欽㹺啾⤥䕕٥♳ꅿ桠㣕٥腉⹲姻➋٥歕歊խ䗼٥虠加խ䗡٥ ✡❁鞢宕٥畾歊꧇䕕 խխ痥畍խـٓ؎ٝسإػٖ٦ءّٝהؐؑ٦ـٖحزח״׷ꦀ誊歗⫷ך涪鋅倜䃊罭♧ խխ痥畍խ鎘皾堣ח״׷猰㷕涸岀⵱٥ٌرٕך涪鋅倯岀ך㾜Ꟛ똲㽵խꥐ٥⯋歊խ嵞 խխ痥畍խ㢳㢌ꆀر٦ةַ׵ך㢳갪䒭㘗岀⵱ך涪鋅⚥ꅿ葺䎂٥俕谏ㄤ䊻 խխ痥畍խ갈㡮ر٦ةك٦أַ׵ך갈㡮濼陎ך涪鋅ꈿ加㛇⛒٥暜ꅿ姻♲ խխ痥畍խ⟎䟝⻉ׁ׸׋➂⡤ַ׵ךشؽ؜٦ءّٝח㛇בֻ濼陎涪鋅ך佄䴂خ٦ٕ뢷谏鞢俑٥둷腜秪♧龤 痥7鿇խطحزٙ٦ؙ橆㞮חֶֽ׷濼陎涪鋅 խխ痥畍խى٦يًر؍،׾欽ְ׋濼頿崧鸐ה猰㷕䪮遭ر٦ةך〳鋔⻉歊⚥խ隁 խխ痥畍խؤ٦ىؚٝ䪮遭׾欽ְ׋㼎鑧涸䞔㜠嗚稊؎ٝةؿؑ٦أ鞢歊姻〷٥叝㿊䝜ㆂ խխ痥畍խؙٔٝ䞔㜠ַ׵ך濼陎笨圓䧭䑜䊛⡟⼪歑٥寑歊㣐鰊٥歊〡ⶌ〷 խխ痥畍խ؎ٝة٦طحزדך⟰噟꟦䞔㜠Ⱏ剣חぢֽ׋وٕثؒ٦آؑٝزءأذي嬁ⵃꥐ㣗٥넝歊酔䘊

⡭锑✳ⴓ寸㹀ؚٓؿהر٦ةو؎صؚٝ 㹧ꅿ䝎⯓欰

⡭锑✳ⴓ寸㹀ؚٓؿהر٦ةو؎صؚٝ ؗ٦ٙ٦س ✳ⴓ寸㹀ؚٓؿ㷕统濼陎涪鋅ػة٦ٝوحثؚٝ،ٕ؞ٔؤير٦ةو؎صؚٝ鎘皾ꆀ䞔㜠ꆀ؜ظير٦ةو؎ٝ؎ٝ ؚ؜ظي䞔㜠灇瑔嚊銲涪鋅׾ׅ׷ֿהכ猰㷕חֶֽ׷剑׮㛇劤涸ז崞⹛ד֮׶ ׉ך涪鋅כ倜׃ְ䔹僤ך涪鋅ד֮׏׋׶ ؛فٓ٦ך岀⵱ך״ֲח錁庠ر٦ة ַ׵㼪ֹ⳿ׁ׸׋暟椚岀⵱ד֮׏׋׶ׅ׷➂꟦ך濼涸崞⹛ה׃גꅾ銲ז⡘縧ח֮׷ֲֿ׃׋濼陎涪鋅כ 剑鵚岣湡ׁ׸גְ׷ر٦ةو
؎صؚٝءأذيך״ֲח ➂꟦ח➿׻׏ג鎘皾堣♳ךءأذيָ遤ֲֿהָ䓼ֻ劍䖉ׁ׸גְ׷ ֲֿ׃׋朐屣ח㼎䘔ׅ׷׋׭ח 劤灇瑔כ ✳ⴓ寸㹀ؚٓؿ #%% ׾濼陎邌植ח欽ְ ر٦ةַ׵ך넝如ך濼陎涪鋅׾〳腉הׅ׷ءأذي ךꟚ涪׾湡涸ה׃ג灇瑔׾Ꟛ㨣׃׋✳ⴓ寸㹀ؚٓؿכ 锷椚㔐騟鏣鎘ךⴓꅿדꅾ銲ז嚊䙀ד֮׶ 锷椚ꟼ侧ךさ䧭剑黝⻉זוד׉ך剣 ⸬䚍ָ涪䳸ׁ׸גְ׷׮ךד֮׷ת׋ ֲֿ׃׋ءأذيךꟚ涪׾鸐׃ג 涪鋅猰㷕ך⾱椚װ椚锷ך圓眠ך㹋鏾涸駈ַָ׶׾אֻ׷ֿה ׮劤灇瑔ך׮ֲ♧אךꅾ銲ז湡涸ד֮׷ 劤灇瑔דכ ת׆寸㹀ؚٓؿ׾濼陎邌植ח欽ְ׷ֿהח֮׷ָ 劤灇瑔׾㾜Ꟛׅ׷⚥ד 鎘皾ꆀחꟼׅ׷灇瑔ַ׵ ׉׸׾湫䱸ח《׶䪔ֲ ֿהחכ㔭ꨇָ⠵ֲֿהָ圫ղך錁挿ַ׵ⴻ僇׃׋׉ך׋׭ 䖞勻ך寸㹀加ה׃גת׆⟎铡׾欰䧭׃ ׉ך䖓寸㹀ت؎،ؚٓي׾箟꧊ ׅ׷הְֲ倯䒭׾ה׏׋ת׋ ⟎铡欰䧭倯䒭ך灇瑔ה׃גכ 圫ղך㷕统倯䒭׾嗚鎢׃ 加圓鸡 ؚٓؿ圓鸡 طحزٙ٦ؙ圓鸡瘝׾㷕统ׅ ׷׋׭ך椚锷涸灇瑔הءأذيꟚ涪׾遤׏׋׉׃גֲֿ׃׋灇瑔׾筨䭍׃׋濼陎涪鋅ءأذيה׃ג)ZQPUIFTJT$SFBUPS׾Ꟛ涪׃ 劤灇瑔דכ⚺ח⟎铡欰䧭ؒٝآٝה׃ג 侧⼪♰㾩䚍ח㼎䘔דֹ׷⚛⴨寸㹀加圓䧭倯䒭׾㹋鄲׃׋剑䖓ח ⟎铡箟꧊ءأذي%FDJTJPO %JBHSBN&EJUPS׾Ꟛ涪ׅ׷ֿהח״׶ )VNBO&YQFSUך➜Ⰵ׾琎噰涸ח佄䴂׃ג 寸㹀加ַ׵寸㹀ت؎،ؚٓي׾⡲׶⳿ׅ׋׭ךءأ ذي׾㸣䧭ׇׁ׋

㣐鋉垷濼陎Ⳣ椚灇瑔㹓傊濼陎ًر؍،灇瑔㹓⻌嵲麣㣐㷕㣐㷕ꤍ䊨㷕灇瑔猰ꨵ㶨䞔㜠䊨㷕㼔余鎘皾堣䞔㜠鸐⥋䊨㷕闌䏟鎘皾堣،٦ؗذؙثٍ䊨㷕ⴓꅿ ⻌嵲麣㣐㷕㣐㷕ꤍ䞔㜠猰㷕灇瑔猰؝ٝؾُ٦ة؟؎ؒٝأ㼔余濼陎اؿزؐؑ،猰㷕闌䏟濼陎ًر؍،灇瑔㹓戣䊛כ⻌㣐㷕欰䔲儗麩ֲ㼔余ך灇瑔㹓̕׌׏׋׋׭ծקר篑ז׃ ⻌嵲麣㣐㷕㣐㷕ꤍ䊨㷕灇瑔猰ءأذي䞔㜠䊨㷕㼔余鎘皾堣䞔㜠鸐⥋䊨㷕闌䏟鎘皾堣،٦ؗذؙثٍ䊨㷕ⴓꅿ

㣐鋉垷濼陎Ⳣ椚灇瑔㹓傊濼陎ًر؍،灇瑔㹓⻌嵲麣㣐㷕㣐㷕ꤍ䞔㜠猰㷕灇瑔猰؝ٝؾُ٦ة؟؎ؒٝأ㼔余濼陎اؿزؐؑ،猰㷕闌䏟濼陎ًر؍،灇瑔㹓⻌嵲麣㣐㷕㣐㷕ꤍ䞔㜠猰㷕灇瑔猰䞔㜠椚䊨㷕㼔余濼陎اؿزؐؑ،猰㷕闌䏟㣐鋉垷濼陎Ⳣ椚灇瑔㹓

㣐鋉垷濼陎Ⳣ椚灇瑔㹓傊濼陎ًر؍،灇瑔㹓

➂䊨濼腉㷕⠓➂䊨濼腉㛇劤㉏겗灇瑔⠓ 4*('1"* େٱอ޷ষ ੴിਖ਼࿨ େୌܒհ খྛ༃໌ ޙ౻ܒհ ຸాྰॹಸ

➂䊨濼腉㷕⠓➂䊨濼腉㛇劤㉏겗灇瑔⠓ 4*('1"* 㷕⠓钞չ➂䊨濼腉պ7PM /P 䎃剢 ؔ٦فٝ،ؙإأ הֿ׹דծ➙剢ך➂䊨濼腉㷕⠓钞ך戣䊛ׁ׿ך4*('1"*鎸✲ծ׋׌ך灇瑔⠓ך稱➜鎸✲ךعؤך זךחծִ׵ְ鋅陎חז׏גְגծ➭ך灇瑔⠓ך鎸✲הך衅䊴ח㔭䞨׃ת׃׋կ׫זׁת׮僽ꬊկ LBTIJ@QPOH⯓欰锷椚㷕统濼陎ך邌植ה栻䖤⚛⴨鎘皾ٌرٕ濼涸فؚٗٓىؚٝ荈搫鎉铂椚鍑ػة٦ٝ椚鍑זוח
ꟼׅ׷➂䊨濼腉ה׃גך㛇燉涸灇瑔剣䊛 https://doi.org/10.11517/jjsai.34.5_603

䌢遤ָֹֽׁ 勞俱؎ٝؿؓ걄㚖 ה暟勞堣圓.** 灇瑔筨䭍䌢遤խ溪぀匌❨㣐㷕㣐㷕ꤍ椚㷕禸灇瑔猰խ侄䱇 걄㚖،سغ؎ؠ ꫬ叛խ䁋぀欵噟䪮遭筨さ灇瑔䨽堣腉勞俱؝ٝؾُذ٦ءّشٕرؠ؎ٝ灇瑔إٝة٦խ筨䭍灇瑔⚺䎌 ⟻谏խ耆
Ⱅ頿鎘皾猰㷕䮶莆頿㔚խث٦ؿ؝٦ر؍ط٦ة 㼗㜥խ薉秀 زْة荈⹛鮦吳⯓畭勞俱䪮遭鿇խث٦ؿفٗؿؑحءّشٕؒٝآص، 㼭靼խ⯋㶨匌⻌㣐㷕勞俱猰㷕넝瘝灇瑔䨽խ侄䱇 ⡟谏խ㻂㶨䞔㜠٥ءأذي灇瑔堣圓խⲤ侄䱇ثُ٦ٔحؼ㣐㷕խ灇瑔㆞ 歊⚥խ⸆ ❨鿪㣐㷕㣐㷕ꤍ䊨㷕灇瑔猰խ勞俱䊨㷕㼔余խ侄䱇濼❨խ鞢酔暟颵٥勞俱灇瑔堣圓窟さ㘗勞俱Ꟛ涪٥䞔㜠㛇湍鿇Ꟍ 䞔㜠窟さ㘗暟颵٥勞俱灇瑔䬿挿խⶰ䬿挿ꞿ 峸歊խ㸩屚匌❨㣐㷕㣐㷕ꤍ倜걄㚖ⶼ䧭猰㷕灇瑔猰խًر؍ٕؕ䞔㜠欰ㄏ㼔余խ侄䱇㼇⦋խ幠⛒ 欵噟䪮遭筨さ灇瑔䨽խせ钙ٔ؟٦ثٍ٦ ⚥❁խ嵞䊻傍珗歊㣐㷕椚䊨㷕遭ꤍխ侄䱇垌〡խ濼⛒ ⚥㣛㣐㷕椚䊨㷕鿇խ侄䱇䔲儗ך窟鎘侧椚灇瑔䨽٥䨽ꞿ

䌢遤ָֹֽׁ 勞俱؎ٝؿؓ걄㚖 ה暟勞堣圓.** https://www.nims.go.jp/MII-I/about/index_g.html ⻌嵲麣㣐㷕խせ钙侄䱇խ歊⚥隁.*ٔ؟٦ث،سغ؎ؠ٦ 匌❨㣐㷕խ侄䱇խ䀤歊溪➂ⶰفٗآؙؑزٔ٦ت٦ 䀤歊倜㷕遭 أػ٦أٌرؚٔٝ חⰕ⹫棳ד⿫⸇

植耵䎃剢傈։䬿挿ךؙٗأ،ه؎ٝز ˖ ⻌嵲麣㣐㷕⻉㷕⿾䘔ⶼ䧭灇瑔䬿挿 81**$3F%% 暴⟣Ⲥ侄䱇 ˖ 椚⻉㷕灇瑔䨽ꬠ倜濼腉窟さ灇瑔إٝة٦ "*1 灇瑔㆞ ˖
ꨄ侔圓鸡٥穈さׇ圓鸡׾⠵ֲ堣唒㷕统 ˖ 䎌稢脄欰暟㷕ך׋׭ך堣唒㷕统稢脄歗⫷ 帾㾴㷕统 ˖ 倜׃ְ،ٕ؞ٔؤي٥剑黝⻉ך㹀䒭⻉ה実鍑岀 ˖ 鷲麓㘗ꨵ㶨겥䗍ꖎ 堣唒㷕统ח״׷⹛涸錁㻊 ˖ 堣唒㷕统ַ׵堣唒涪鋅פ ˖ ⻉㷕⿾䘔ךرؠ؎ٝה涪鋅ך׋׭ך堣唒㷕统 ˖ ⴓ㶨ךؚٓؿ邌植ך㷕统ה欰䧭 ˖ ꆀ㶨⻉㷕鎘皾堣唒㷕统ך輐さ ˖ 堣唒涪鋅䱱稊ծ㹋꿀鎘歗ծ濼陎涪鋅

椚⻉㷕灇瑔䨽ꬠ倜濼腉窟さ灇瑔إٝة٦ 東京駅理研AIP 神田ラボ皇居 $03&%0傈劤堀'

⹶⹡㖑❨ꢻ㣽㖑⼒ ❨鿪䏍湱嚂龾礵螟歕 https://www.kobe.riken.jp/about/map/keihanna/ 痥♲إؙة٦ך灇瑔䨽

⹶⹡㖑❨ꢻ㣽㖑⼒ ❨鿪䏍湱嚂龾礵螟歕

⹶⹡㖑❨ꢻ㣽㖑⼒ ❨鿪䏍湱嚂龾礵螟歕 ࠃཱࠃձਤॻؗؔ੢ؗ 㕂ꥷꨵ孡鸐⥋㛇燉䪮遭灇瑔䨽

㕂ꥷꨵ孡鸐⥋㛇燉䪮遭灇瑔䨽"53 瀖랲暴灇ך ،ٝسٗ؎س չؒٔؕպ圫ָꔨ䏟㹋暟כ乆䕦犜姺 ˖ 椚⻉㷕灇瑔䨽 ꬠ倜濼腉窟さ灇瑔إٝة٦"*1 ؖ٦ر؍،ٝٗنحزفٗآؙؑز(31 ˖
"53膷䞔㜠鸐⥋筨さ灇瑔䨽膷䞔㜠灇瑔䨽钠濼堣圓灇瑔䨽膷䞔㜠鍑匿灇瑔䨽 https://www.atr.jp ˖ "53帾㾴؎ٝةؙٓءّٝ筨さ灇瑔䨽 ؎ٝةؙٓءّٝ䪮遭غؙٝ ؎ٝةؙٓءّٝ猰㷕灇瑔䨽瀖랲嵞暴ⴽ灇瑔䨽蠗歊秀⽆暴ⴽ灇瑔䨽 ˖ "53黝䘔؝ىُص؛٦ءّٝ灇瑔䨽 ˖ "53岚⹛䊨㷕灇瑔䨽 ˖ "53⡟谏⻟䗗暴ⴽ灇瑔䨽

椚灇"*1!"53 ˖ ꣇拄猰㷕ث٦ي ♳歊⥜⸆ ˖ 膷䞔㜠窟さ鍑匿ث٦ي 䊛ꑚ♧儙 ˖
鎘皾膷ت؎شىؙأث٦ي 㿊♴㸾➂ ˖ J14稢脄鸬䵿⼔㷕涸ٔأؙ㔐鼘ث٦ي ♳歊⥜⸆ 椚灇"*1ה❨㣐J14稢脄灇ך鸬䵿ٓن

俑猰満⚅歲زحفٖكٕ䬿挿䕎䧭فؚٗٓي 81* https://www.mext.go.jp/a_menu/kagaku/toplevel/ ✮皾鋉垷剑㣐⭙ⰼ䎃䎃

俑猰満⚅歲زحفٖكٕ䬿挿䕎䧭فؚٗٓي 81* ر٦ة٥䞔㜠猰㷕חꅾז׷ 81*䬿挿כ✳䬿挿ך׫ صُ٦ٗ؎ٝذٔآؑٝأ㕂ꥷ灇瑔堣圓⻌㣐䬿挿כչ⻉㷕⿾䘔ךرؠ؎ٝה涪鋅պ

⻌㣐81*䬿挿⻉㷕⿾䘔ⶼ䧭灇瑔䬿挿 *$3F%% ⻉㷕⿾䘔ה䞔㜠猰㷕

⻌㣐81*䬿挿⻉㷕⿾䘔ⶼ䧭灇瑔䬿挿 *$3F%% https://www.icredd.hokudai.ac.jp 䎃䏝 ظ٦كٕ⻉㷕颣「颣

剑㼭ꣲך⵸ֶֹ⻉㷕暟颵"׾#ח㢌ִ׋ְ ⾱㶨ָמהתהת׶חז׏׋չⴓ㶨պה׉ך⻉㷕⿾䘔כ䎢眔㔲זⴓꅿך⚺➂Ⱅ ⻉㷕ծ⼔㷕٥欰椚㷕ծ暟椚㷕ծ欰暟㷕ծⶼ讒ծ勞俱猰㷕ծ橆㞮猰㷕ծ鴋㷕ծ굸ㅷծ⻉礅ㅷծ˘ 魦ך㔐׶ך׮ך 猘׋׍荈魦欰ㄏ植韋 ؒطؘٕ٦ ⻉㷕

ⴓ㶨כչ穈さׇ涸պ⩎꬗׾䭯א https://cen.acs.org/physical-chemistry/computational-chemistry/Exploring-chemical-space-AI-take/98/i13

堣唒㷕统ד⻉㷕׾װ׏ג׫׋ ⴓ㶨כ⾱㶨꟦ך穠さך穈さׇ穈さׇ鋉⵱כꆀ㶨⸂㷕̔䪮遭♳ךꟼ䗰כչ穈さׇך堣唒㷕统պ ⻉㷕⿾䘔כ穠さך穈׫剏ִ 俑岀ח⡂גְ׷ 銲稆ך穈さׇח圓䧭䚍הꥡ㾴䚍ָ֮׷ 鎉铂ח⡂גְ׷

堣唒㷕统ד⻉㷕׾װ׏ג׫׋ 劤䔲ך؞٦ٕכչ⻉㷕⿾䘔ךرؠ؎ٝה涪鋅պד堣唒㷕统כ׉ך׋׭ך麣Ⱗך♧א $挥⻉宏稆겲ךさ䧭 ًةٝ さ䧭ؖأ $0 ) $ ) )
) ) 㣓搫ؖأ٥ؔ؎ٕءؑ٦ٕ٥瀖挥٥غ؎ؔوأזו瀖屘⟃㢩ַ׵䖤׵׸׷ $⻉㷕 $0ך项彁ⵃ欽➂䊨⯔さ䧭⾱俱כ$ָ♧א ⴓ㶨כ⾱㶨꟦ך穠さך穈さׇ穈さׇ鋉⵱כꆀ㶨⸂㷕̔䪮遭♳ךꟼ䗰כչ穈さׇך堣唒㷕统պ ⻉㷕⿾䘔כ穠さך穈׫剏ִ 俑岀ח⡂גְ׷ 銲稆ך穈さׇח圓䧭䚍הꥡ㾴䚍ָ֮׷ 鎉铂ח⡂גְ׷

堣唒㷕统ך⸂ך⯔僇꬗ה农랲꬗ 2堣唒㷕统٥堣唒涪鋅ך䪮遭׏ג荈搫植韋ך椚鍑٥涪鋅ך䕵ח甧אךַז Ԩ ⴓ㶨ך邌植㷕统ה(SBQI/FVSBM/FUXPSLT Ԩ 䌓秛غ؎،أך鏣鎘הؚٖ؎نحؙأ剑黝⻉ ٓ؎ز؟؎س⯔僇꬗ ت٦ؙ؟؎س农랲꬗ Ԩ 繎欰Ꟍ⸬卓ה6OEFSTQFDJDBUJPO
Ԩ չ✮庠ָדֹ׷ֿהպכչ椚鍑պװչ涪鋅պָדֹ׷ֿה׾䠐㄂׃זְ May the ML force be with you… 堣唒㷕统כչر٦ة׾✮庠ח㢌ִ׷պ䓼⸂זذؙظٗآ٦ 荈搫猰㷕ך㹋植韋ر٦ةד⢪ֲךכְ׹ְ׹慧يؤ ؿؓ٦أ ؿؓ٦أהⰟח֮׵׿ֿה׾

❨㣐䎃䏝濼腉䞔㜠㷕㼔余؝ٗؗؐي https://kuist-colloquium.github.io ⹛歗 IUUQTWJNFPDPN

堣唒㷕统כչ倜׃ְ ꧟ז ؝ٝؾُ٦ةفؚٗٓيך⡲׶倯պ ♧菙暟⡤钠陎 ؜٦يفٖ؎ ˑ֮׶ָהֲ˒ J’aime la musique I
love music 갈㡮钠陎堣唒缺鏬馄鍑⫷ فؚٗٓيךⰅ⸂ה⳿⸂ךꟼ⤘ָ״ֻⴓַ׵זְ㜥さד׮ծ ׋ֻׁ׿ךⰅ⳿⸂ך鋅劤ر٦ةח״׏ג꟦䱸涸ח鋅劤׾ⱄ植ׅ׷فؚٗٓي׾⡲׶⳿ׅ䪮遭

堣唒㷕统כչ倜׃ְ ꧟ז ؝ٝؾُ٦ةفؚٗٓيך⡲׶倯պ ֿך⽃秪ז׃ֻ׫כ♳䩛ח⢪ֲהչ׭׍ׯֻ׍ׯ䓼⸂պדְ׹ְ׹ז嚂׃ְֿה׮דֹ׷

➬穈׫כ刼꬗ؿ؍حذ؍ؚٝח״׷ر٦ةⰻ䯏 չ刼꬗ٌرٕպךⰻ鿇ػًٓة⦼׾锃侭׃ג鋅劤挿חֲ֮״ֲؿ؍حذ؍ؚׅٝ׷ Random Forest Neural Network SVR Kernel Ridge p1
p2 p3 p4  刼꬗ٌرٕ

⼱⚅秀⵸ח欰ת׸׋չ䋞劄涸זㄎ獥պח״׷䎐䟝ח׀岣䠐植㖈ך堣唒㷕统כ♧菙ח䟝⫷ׅ׷4'涸זչ➂䊨濼腉 "* պהכַז׶ַֽꨄ׸גְ׷ָծ չر٦ة׾✮庠ח㢌ִ׷պ堣腉ָ֮ת׶ח䓼⸂ז׋׭ծ猘׋׍ך傈䌢欰崞ַ׵➙䖓ך爡⠓ך ؕةثחתד䕦갟׾⿹ר׉ֲה׃גְ׷˘ չ➂䊨濼腉պչ堣唒㷕统պזוך䋞劄涸זㄎ獥כ劤颵׾ىأٔ٦س׃װְׅךד׀岣䠐׾ wishful mnemonics Why
Is AI So Dumb? A SPECIAL REPORT https://spectrum.ieee.org/ﬁles/11920/10_Spectrum_2021.pdf

-FP#SFJNBOךٖحأٝ3BTIPNPO 0DDBN #FMMNBO https://projecteuclid.org/euclid.ss/1009213726 (Open Access) ˖ 6$#FSLFMFZせ钙侄䱇 ˖ 4*(,%%*OOPWBUJPO"XBSE
˖ 1SPCBCJMJUZ5IFPSJTU̔ $POTVMUBOU̔4UBUJTUJUJBO -FP#SFJNBO

The Data Modeling Culture The Algorithmic Modeling Culture -FP#SFJNBOךٖحأٝ3BTIPNPO 0DDBN
#FMMNBO ˖ 寸㹀加،ٝ؟ٝـٕ ˖ 荈⹛䗍ⴓؿٖ٦يٙ٦ؙ 帾㾴㷕统

-FP#SFJNBOךٖحأٝ3BTIPNPO 0DDBN #FMMNBO

➙ד׮ֿך挿כ0QFOז㉏겗׾㶱׿ד堣唒㷕统ך⚥䗰㉏겗ה׃ג灇瑔ׁ׸גְ׷ Ӎ "Rashomon"葺ְٌرٕך㢳ꅾ䚍 ꬊ♧䠐䚍ず玎䏝ך葺ְ✮庠礵䏝׾䭯אⰋֻ殯ז׷ٌرָٕ׋ֻׁ׿㶷㖈ׅ׷ Ӎ "Occam"ٌرٕך鍑ꅸ䚍ה✮庠礵䏝ך؝ٝؿؙٔز ٌرٕךءٝفׁٕ 鍑ꅸ䚍
ה✮庠礵䏝ך⚕甧כהג׮ꨇ׃ְ Ӎ "Bellman"넝如⯋ر٦ةָ䒷ֹ饯ֿًׅٔحزהرًٔحز 넝如⯋ז邌植 ꟼ⤘׃׉ֲזדֹ׷׌ֽ㢳ֻך㢌ꆀ ׾䪔ֲץֹזךַ ⠗窟涸ז窟鎘㷕ך״ֲח佄ꂁ涸ז㼰侧ך㢌ꆀ׾嗚鎢׃ⴓ匿ׅץֹזךַ ̔ؗحثٝءؙٝ㔐䌓䙼ְאֻ㢌侧Ⰻ鿇Ⰵ׶ٌرٕ ٥暴䗙ꆀ ؒٝآٔ،ؚٔٝה亻⡂湱ꟼ٥3BTIPNPO⸬卓ך㟓㣐ٔأؙ 3BTIPNPO 0DDBN #FMMNBO -FP#SFJNBOךٖحأٝ3BTIPNPO 0DDBN #FMMNBO

帾㾴㷕统ה邌植㷕统 q1 q2 q3 q4  刼꬗ٌرٕ Ⰵ⸂邌植垥彊涸ז 堣唒㷕统
ر٦ةⰻ䯏 Random Forest GBDT Nearest Neighbor SVM Gaussian Process Neural Network deep learning

帾㾴㷕统ה邌植㷕统 q1 q2 q3 q4  刼꬗ٌرٕ p1 p2 p3
p4  㢌侧㢌䳔邌植㷕统 <latexit sha1_base64="wThlqV4N/f6TJkASaopeeXy6hGk=">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</latexit> z1 <latexit sha1_base64="eRVZ9lBHtuomnSZ9U9icTCKNkZM=">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</latexit> z2 Ⰵ⸂邌植 deep learning

p4  㢌侧㢌䳔邌植㷕统 <latexit sha1_base64="wThlqV4N/f6TJkASaopeeXy6hGk=">AAAChnichVHLTsJAFD3UF+ID1I2JGyLBuCJTg2JcEd245CGPBAlp64gNpW3aQgLEHzBxKwtXmrgwfoAf4MYfcMEnGJeYuHHhpTQxSsTbTOfMmXvunJkrm5pqO4z1fMLE5NT0jH82MDe/sBgMLS3nbaNhKTynGJphFWXJ5pqq85yjOhovmhaX6rLGC3LtYLBfaHLLVg39yGmZvFyXqrp6qiqSQ1S2XREroQiLMTfCo0D0QARepIzQI45xAgMKGqiDQ4dDWIMEm74SRDCYxJXRIc4ipLr7HOcIkLZBWZwyJGJr9K/SquSxOq0HNW1XrdApGg2LlGFE2Qu7Z332zB7YK/v8s1bHrTHw0qJZHmq5WQlerGY//lXVaXZw9q0a69nBKXZdryp5N11mcAtlqG+2u/3sXiba2WC37I3837Aee6Ib6M135S7NM9dj/MjkhV6MGiT+bscoyG/FxJ1YPB2PJPe9VvmxhnVsUj8SSOIQKeSofhWXuEJX8AsxYVtIDFMFn6dZwY8Qkl9Y9pCe</latexit> z1 <latexit sha1_base64="eRVZ9lBHtuomnSZ9U9icTCKNkZM=">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</latexit> z2 Ⰵ⸂邌植 <latexit sha1_base64="wThlqV4N/f6TJkASaopeeXy6hGk=">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</latexit> z1 <latexit sha1_base64="eRVZ9lBHtuomnSZ9U9icTCKNkZM=">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</latexit> z2 葺ְ邌植 deep learning

p4  㢌侧㢌䳔邌植㷕统 <latexit sha1_base64="wThlqV4N/f6TJkASaopeeXy6hGk=">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</latexit> z1 <latexit sha1_base64="eRVZ9lBHtuomnSZ9U9icTCKNkZM=">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</latexit> z2 Ⰵ⸂邌植 <latexit sha1_base64="wThlqV4N/f6TJkASaopeeXy6hGk=">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</latexit> z1 <latexit sha1_base64="eRVZ9lBHtuomnSZ9U9icTCKNkZM=">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</latexit> z2 葺ְ邌植 ءٝفٕ 简䕎 ד⼧ⴓ 垥彊涸ז 堣唒㷕统 ر٦ةⰻ䯏 deep learning

p4  㢌侧㢌䳔邌植㷕统 <latexit sha1_base64="wThlqV4N/f6TJkASaopeeXy6hGk=">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</latexit> z1 <latexit sha1_base64="eRVZ9lBHtuomnSZ9U9icTCKNkZM=">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</latexit> z2 Ⰵ⸂邌植 <latexit sha1_base64="wThlqV4N/f6TJkASaopeeXy6hGk=">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</latexit> z1 <latexit sha1_base64="eRVZ9lBHtuomnSZ9U9icTCKNkZM=">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</latexit> z2 葺ְ邌植帾㾴㷕统 ءٝفٕ 简䕎 ד⼧ⴓ 垥彊涸ז 堣唒㷕统 ر٦ةⰻ䯏 deep learning

p4  㢌侧㢌䳔邌植㷕统 <latexit sha1_base64="wThlqV4N/f6TJkASaopeeXy6hGk=">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</latexit> z1 <latexit sha1_base64="eRVZ9lBHtuomnSZ9U9icTCKNkZM=">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</latexit> z2 Ⰵ⸂邌植 <latexit sha1_base64="wThlqV4N/f6TJkASaopeeXy6hGk=">AAAChnichVHLTsJAFD3UF+ID1I2JGyLBuCJTg2JcEd245CGPBAlp64gNpW3aQgLEHzBxKwtXmrgwfoAf4MYfcMEnGJeYuHHhpTQxSsTbTOfMmXvunJkrm5pqO4z1fMLE5NT0jH82MDe/sBgMLS3nbaNhKTynGJphFWXJ5pqq85yjOhovmhaX6rLGC3LtYLBfaHLLVg39yGmZvFyXqrp6qiqSQ1S2XREroQiLMTfCo0D0QARepIzQI45xAgMKGqiDQ4dDWIMEm74SRDCYxJXRIc4ipLr7HOcIkLZBWZwyJGJr9K/SquSxOq0HNW1XrdApGg2LlGFE2Qu7Z332zB7YK/v8s1bHrTHw0qJZHmq5WQlerGY//lXVaXZw9q0a69nBKXZdryp5N11mcAtlqG+2u/3sXiba2WC37I3837Aee6Ib6M135S7NM9dj/MjkhV6MGiT+bscoyG/FxJ1YPB2PJPe9VvmxhnVsUj8SSOIQKeSofhWXuEJX8AsxYVtIDFMFn6dZwY8Qkl9Y9pCe</latexit> z1 <latexit sha1_base64="eRVZ9lBHtuomnSZ9U9icTCKNkZM=">AAAChnichVHLTsJAFD3UF+ID1I2JGyLBuCIDQTGuiG5c8pBHgoS0dcCG0jZtIQHiD5i4lYUrTVwYP8APcOMPuOATjEtM3LjwUpoYJeJtpnPmzD13zsyVDFWxbMb6HmFqemZ2zjvvW1hcWvYHVlbzlt40ZZ6TdVU3i5JocVXReM5WbJUXDZOLDUnlBal+ONwvtLhpKbp2bLcNXm6INU2pKrJoE5XtVGKVQIhFmBPBcRB1QQhupPTAI05wCh0ymmiAQ4NNWIUIi74SomAwiCujS5xJSHH2Oc7hI22TsjhliMTW6V+jVcllNVoPa1qOWqZTVBomKYMIsxd2zwbsmT2wV/b5Z62uU2PopU2zNNJyo+K/WM9+/Ktq0Gzj7Fs10bONKvYcrwp5NxxmeAt5pG91eoPsfibc3WK37I3837A+e6IbaK13+S7NM9cT/EjkhV6MGhT93Y5xkI9ForuReDoeSh64rfJiA5vYpn4kkMQRUshR/RoucYWe4BUiwo6QGKUKHlezhh8hJL8AWxaQnw==</latexit> z2 葺ְ邌植帾㾴㷕统 ءٝفٕ 简䕎 ד⼧ⴓ 垥彊涸ז 堣唒㷕统 ر٦ةⰻ䯏 deep learning ر٦ةⰻ䯏כⰅ⸂邌植דכזֻ չ葺ְ邌植պד遤ֲ

䗍ⴓ〳腉فؚٗٓىؚٝ ... backbone head https://en.wikipedia.org/wiki/Differentiable_programming さ䧭ꟼ侧׾鎘皾ؚٓؿד邌植ׅ׸ל⺟ꂁכ荈⹛䗍ⴓ BLBCBDLQSPQ ד鎘皾〳̔⺟ꂁ岀ד剑黝⻉ + concat
× softmax "All numerical computations are ultimately compositions of a ﬁnite set of elementary operations for which derivatives are known (Verma, 2000; Griewank and Walther, 2008)" Automatic differentiation in machine learning: a survey, JMLR 18 (2018) 1-43 さ䧭ך➬倯כ荈歋

岣䠐堣圓ה圓䧭涸㷕统 넝䏝זةأؙךⳢ椚׾圓䧭銲稆٥ٌآُ٦ٕה׉ך穈さׇさ䧭ח״׶㹋植 ⢽ ˖ ⽃铂׾ֲתֻ穈׫さ׻ׇ׷ֿהד圫ղז䠐㄂ך俑畍׾搀侧ח⡲׶⳿ׅ ˖ 圫ղזٌآُ٦ٕ׾ֲתֻ穈׫さ׻ׇ׷ֿהד膷ך넝如堣腉׾厫鮾ח⡲׶⳿ׅ 醱侧ךⰅ⸂
Key Query Value 岣䠐堣圓כ鋙ִגְֶ׋䞔㜠׾DPOUFYUח䘔ׄגה׶׌ׅ 鳤剅ؔـآؙؑز RVFSZ LFZ WBMVF ה׫זׇ׷ ˖ 俑腞חꟼ׃ג⾨ְֿה $POUFYUUIJDL 㢳圫ז珏겲ך俑腞חְֶג黝ⴖז⹛⡲׾遤ֲ腉⸂ ˖ ꅾ銲ז䞔㜠חְַח׃ג岣䠐 "UUFOUJPO ׾ぢֽ׷ַ 넝䏝זةأؙח䗳銲俑腞⣛㶷涸ז 䞔㜠ך铣׫⳿׃ ꥡ㾴涸堣圓 ⚛遤涸堣圓㢳圫ז تؐٝ أزٔ٦ي ةأؙ

㔭ꨇ΍植➿ך堣唒㷕统ٌرٕכ㢳ꆀךر٦ة׾䗳銲הׅ׷ q1 q2 q3 q4  刼꬗ٌرٕ p1 p2 p3
p4  㢌侧㢌䳔邌植㷕统 <latexit sha1_base64="mpuRLmJhhTRKlPt+QsNIZJRw5HU=">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</latexit> x1 <latexit sha1_base64="C1ohMOVeKO2tnLKPo8dfEeHHC5o=">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</latexit> x2 <latexit sha1_base64="80tKy5O2X9pRix5cNVQWm3Mk5KI=">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</latexit> xd  ΍넝如⯋䚍Ⰵ⸂㢌侧ָ㢳ֺׅ Ύ麓ⶱػًٓة⻉ػًٓة侧ָ㢳ֺׅ 3FT/FU♰ػًٓة 3FT/FU♰ػًٓة &DJFOU/FU#♰ػًٓة 7((⭙♰ػًٓة 歗⫷׉ךתת׾Ⰵ⸂ׅ׷㜥さ ؾؙإٕךؕٓ٦歗⫷̔㢌侧 ؾؙإٕךؕٓ٦歗⫷̔♰㢌侧 MBZFS IFBET#&35⭙♰ػًٓة MBZFS IFBET#&35⭙♰ػًٓة (159-⭙♰ػًٓة (15⭙ػًٓة Ԩ 堣唒㷕统כⰅ⸂ׁ׸גזְ䞔㜠׾Ⰻֻ 罋䣁׃גֻ׸זְ˘ 亻⡂湱ꟼٔأؙ Ԩ ה׶ִ֮׆葿ղז㢌侧׾Ⰵ׸ָ׍ 歗⫷ 鎉铂

㔭ꨇΎ繎欰Ꟍ⸬卓ה6OEFSTQFDJGJDBUJPO 繎欰Ꟍ⸬卓葺ְ堣唒㷕统ٌرٕך㢳ꅾ䚍 ꬊ♧䠐䚍 넝ְ✮庠礵䏝׾䭯א堣唒㷕统ٌرٕכ♧אךر٦ةإحزַ׵׋ֻׁ׿⡲׸׷ 剣ꣲر٦ةַ׵鋅琎׮׷✮庠礵䏝דכ䭷侧涸䎢ׁךٌرٕ䱱稊瑞꟦׾TQFDJGZ׃ֹ׸זְ˘

㔭ꨇΎ繎欰Ꟍ⸬卓ה6OEFSTQFDJGJDBUJPO 繎欰Ꟍ⸬卓葺ְ堣唒㷕统ٌرٕך㢳ꅾ䚍 ꬊ♧䠐䚍 넝ְ✮庠礵䏝׾䭯א堣唒㷕统ٌرٕכ♧אךر٦ةإحزַ׵׋ֻׁ׿⡲׸׷ 剣ꣲر٦ةַ׵鋅琎׮׷✮庠礵䏝דכ䭷侧涸䎢ׁךٌرٕ䱱稊瑞꟦׾TQFDJGZ׃ֹ׸זְ˘ Ԩ չוךٌرָٕ実׭׷溪㹋זךպה罋ִג׃תֲהծתׁח溪㹋כչ谞ך⚥պ˘ ̔醱侧ך䩛岀ח״׷㢳錬涸鍑ꅸָꉟ⵱ "all models
are wrong but some models are useful" Ԩ 㹋ꥷחכ劤颵涸חر٦ةָ駈׶גזְ 6OEFSTQFDJDBUJPO ֿהד㢳ꅾ䚍כׁ׵ח䝤⻉ ׌ְ׋ְך倯岀ד겲⡂ 䩛岀װٌرٕח״׏ג✮庠儗ך䮙⹛חַז׶䊴ָ⳿ג׃תֲ

6OEFSTQFDJGJDBUJPOכؽحؚر٦ة✲⢽ד׮䎢ֻ饯ֿ׏גְ׷ https://arxiv.org/abs/2011.03395 https://ai.googleblog.com/2021/10/ how-underspeciﬁcation-presents.html ءّنְ钠濼腉⸂ךֶתִ׵➂꟦חה׏׋׵չؽحؚպر٦ةַ׮׃׵׿ֽוծ مٝوח䗳銲ז䞔㜠ꆀַ׵׃׋׵عشؙا׫׋ְז׮׿װזխ CZر؍٦فٓ٦صؚٝ圫

植➿ך䪮遭涸ꟼ䗰כֿך넝如⯋䚍׾וֲ䩛䥿ֽ׷ַ 然桦涸剑黝⻉٥姻⵱⻉ ✲⵸㷕统 8BSN4UBSU ך鯄獳 ٌرָٕ㣐ְֹ荈歋䏝ך⚥ד凐׸תֻ׵זְ״ֲ ⹛ֽ׷眔㔲׾⡦הַ׃גⵖ䖴٥ⵖꣲ٥㸜㹀⻉ׅ׷ ⻉㷕涸ח㧅䔲䚍׾妀ֻ״ֲזٌرָٕ䠐㔳ׇ׆邌植ׁ׸ג
׃ת׻זְ״ֲח⻉㷕ך濼陎װ椚锷猰㷕٥鎘皾⻉㷕ך濼鋅׾ 筨⹛㆞׃גٌرٕך荈歋䏝׾䪮遭涸חⵖꣲׅ׷ 堣唒㷕统⻉㷕⻉㷕ח黝さ׃׋䌓秛غ؎،أךرؠ؎ٝ p1 p2 p3 p4  刼꬗ٌرٕ 刼꬗ٌرָٕו׿זⰅ⳿⸂ꟼ⤘ד׮邌植דֹ׷ֿהָ鷞ח 亻⡂湱ꟼװ6OEFSTQFDJDBUJPOך㉏겗׾䝤⻉ׇׁגְ׷ 䌓秛غ؎،أך鏣鎘 ✲⵸ח䖤גְֶ׋؎؎䠬ׄךػًٓةⴱ劍⦼׾⢪ֲ

㹋⢽ⴓ㶨ך邌植㷕统ה(SBQI/FVSBM/FUXPSLT (//T زهٗآ ⴓ㶨٥勞俱ך橆㞮勴⟝垥涸湱✼⡲欽˘ ؚٓؿ邌植鴟暴䗙 갥挿暴䗙 p1 p2 p3
 邌植㷕统 Ⰵ⸂䞔㜠ךさ䧭怴皾 q1 q2 q3  刼꬗ٌرٕ

䎗⡦涸帾㾴㷕统 https://arxiv.org/abs/2104.13478 https://youtu.be/uF53xsT7mjc https://youtu.be/w6Pw4MOzMuo ICLR 2021 Keynote (Michael Bronstein) Seminar
Talk (Petar Veličković) (//כ䌴䎢ְ䎗⡦圓鸡׾窟♧涸ח䪔ִ׷単穈׫ 堣唒㷕统ךٕؒٓٝ؜ٝ٥فؚٗٓي

堣唒㷕统⻉㷕ך溪ך㉏겗 ♴鎸כו׸׮堣唒㷕统דַז׶넝礵䏝ז✮庠ָדֹתָׅծ׉׸כ猘׋׍ָ׉ך➬穈׫׾ 椚鍑דֹ׋ֿה׾㼰׃ד׮䠐㄂ׅ׷ד׃׳ֲַ וֲװ׏ג鋅ִגְ׷׮ך׾钠陎 וֲװ׏׋׵㔲烝⹧ג׷ ˑ֮׶ָהֲ˒ J’aime la musique I
love music וֲװ׏ג갈岚ַ׵鎉衝׾椚鍑 וֲװ׏ג㢳鎉铂׾椚鍑 չ✮庠ָדֹ׷պֿהכչ椚鍑պװչ涪鋅պָדֹ׷ֿה׾湫䱸כ䠐㄂׃זְ

㔓卓ך椚鍑חכ㹋꿀灇瑔 ➜Ⰵ灇瑔 ָ䗳銲♶〳妀堣唒㷕统כֻ֮תדر٦ةך⚥ך㢳如⯋湱ꟼ׾䯝ִծ׉׸ח״׏ג✮庠ׅ׷䪮遭 Ը錁㻊ׁ׸׋湱ꟼָ劤䔲ח㔓卓䚍׾ろ׬ךַ׾然ַ׭׷׋׭חכ㹋꿀ׅ׷׃ַזְ ⡤ꅾ 魦ꞿ 傈劤فٗꅿ椔Ꟛ䍊♧鮩鼅䩛ך魦ꞿ٥⡤ꅾر٦ة 䎃椔㔚Ⱅ䒭؟؎ز鼅䩛ر٦ة״׶荈⡲
չ⡤ꅾ׾㟓װׇל魦ꞿ׮⠼ן׷պָ姻׃ְַכ ֿך錁㻊ر٦ة׌ַֽ׵כ寸׃גⴓַ׵זְ 䘔欽窟鎘㷕ך㛇劤ךؗ 湱ꟼꟼ⤘כ䗳׆׃׮㔓卓ꟼ⤘׾䠐㄂׃זְ չ✮庠ָדֹ׷պֿהכչ椚鍑պװչ涪鋅պָ דֹ׷ֿה׾湫䱸כ䠐㄂׃זְ

✮庠ַ׵椚鍑٥涪鋅פ걧ד׏ַ׍׾膴׃ծ植㜥ח⳿״ֲ ✲⟝כ؝ٝؾُ٦ة 堣唒㷕统 ך⚥ד饯ֹג׷׿ׄׯזְծ植㜥ד饯ֹגְ׷׿׌CZ⥯ 邌植 ➜Ⰵ ˖ 剣⸬זⰅ⸂㢌侧ךず㹀 ˖ 堣唒㷕统ةأؙךرؠ؎ٝ
˖ 鎮箺ر٦ةך鏣鎘ה　꧊ ˖ 傀㶷ךر٦ة٥濼陎ךⵃ欽 ˖ 㹋꿀ך鎘歗ה㹋倵 ˖ 穠卓ך鐰⣣ה鍑ꅸ ˖ 㹋꿀⻉㷕罏הך⼿⫴ ˖ 椚锷⻉㷕罏הך⼿⫴ 堣唒㷕统ٌرٕ 植㜥 ؝ٝؾُ٦ةך⚥ 猘׋׍װ㹋植韋植㜥ך䞔㜠׾Ⰵ׸׷ 䞔㜠׾植㜥ח䨱ׅ

侄鎮猰㷕灇瑔הכ穠㽷➂꟦ך㌀׫ չ椚鍑պװչ涪鋅պ׃׋ְךכ堣唒דכזֻ猘׋׍➂꟦ את׶ծ荈搫岀⵱ך㉏겗דכזֻ猘׋׍荈魦ך礵牞ה⚅歲ך֮׶倯ך㉏겗׾㉏ֲֿהחז׷

侄鎮猰㷕灇瑔הכ穠㽷➂꟦ך㌀׫ չ椚鍑պװչ涪鋅պ׃׋ְךכ堣唒דכזֻ猘׋׍➂꟦ את׶ծ荈搫岀⵱ך㉏겗דכזֻ猘׋׍荈魦ך礵牞ה⚅歲ך֮׶倯ך㉏겗׾㉏ֲֿהחז׷ ˖ 鍑ꅸ䚍猘׋׍ךءّنְ钠濼腉⸂ח　ת׷״ֲזչ䎂僒ז椚鍑պָ実׭׵׸גְ׷կ

侄鎮猰㷕灇瑔הכ穠㽷➂꟦ך㌀׫ չ椚鍑պװչ涪鋅պ׃׋ְךכ堣唒דכזֻ猘׋׍➂꟦ את׶ծ荈搫岀⵱ך㉏겗דכזֻ猘׋׍荈魦ך礵牞ה⚅歲ך֮׶倯ך㉏겗׾㉏ֲֿהחז׷ ˖ 鍑ꅸ䚍猘׋׍ךءّنְ钠濼腉⸂ח　ת׷״ֲזչ䎂僒ז椚鍑պָ実׭׵׸גְ׷կ ˖ 儗꟦䚍搀ꣲח儗꟦ָ֮׸לչ㷕统պכ♶銲կ剣ꣲך儗꟦׃ַ欰ֹ׵׸זְ猘׋׍חչ涪鋅պ ׾ֶ苇甧גׅ׷׋׭ךؼٝز⳿׃ָ実׭׵׸גְ׷ ➂겲窫徦ךة؎ئىحزⰻח կ

˖ 䞔㜠ך鿇ⴓ䚍ر٦ةחדֹ׷䞔㜠כְאד׮⚅歲ך䞔㜠ꆀךק׿ךמהַֽ׵׌ֽկ ײֻ屎ך崧׸כ窫ִ׆׃גծ׃ַ׮׮הך宏ח֮׵׆կׅץג׾錁庠ׅ׷ֿהכדֹזְկ

˖ 䞔㜠ך鿇ⴓ䚍ر٦ةחדֹ׷䞔㜠כְאד׮⚅歲ך䞔㜠ꆀךק׿ךמהַֽ׵׌ֽկ ײֻ屎ך崧׸כ窫ִ׆׃גծ׃ַ׮׮הך宏ח֮׵׆կׅץג׾錁庠ׅ׷ֿהכדֹזְկ ˖ 錁庠غ؎،أ➂꟦ָ♧欰䧄ㄏ꧊׭׋ر٦ةכוֲ׃׋׏ג⡦׵ַך⨉׶ַ׵鷕׸׵׸זְկ ♷ִ׵׸׋ر٦ةך⫘ぢ׾䯝ִ׷堣唒㷕统ך✮庠׮ず圫ח׉ך⨉׶ַ׵鷕׸׵׸זְկ

˖ 䞔㜠ך鿇ⴓ䚍ر٦ةחדֹ׷䞔㜠כְאד׮⚅歲ך䞔㜠ꆀךק׿ךמהַֽ׵׌ֽկ ײֻ屎ך崧׸כ窫ִ׆׃גծ׃ַ׮׮הך宏ח֮׵׆կׅץג׾錁庠ׅ׷ֿהכדֹזְկ ˖ 錁庠غ؎،أ➂꟦ָ♧欰䧄ㄏ꧊׭׋ر٦ةכוֲ׃׋׏ג⡦׵ַך⨉׶ַ׵鷕׸׵׸זְկ ♷ִ׵׸׋ر٦ةך⫘ぢ׾䯝ִ׷堣唒㷕统ך✮庠׮ず圫ח׉ך⨉׶ַ׵鷕׸׵׸זְկ ˖ 㔓卓䚍ך椚鍑չ㔓卓䚍պכ湫䱸錁庠דֹזְկ➂꟦ָ،ؙإأדֹ׷ךכչ湱ꟼպ׌ֽկ

堣唒㷕统ַ׵堣唒涪鋅פ Science 363 (2019) Nature 583 (2020) Nature Reviews Drug
Discovery 17 (2018) 㹋꿀荈⹛⻉ך䪮遭涸涪㾜⻉㷕ד׮ꬊ⸬桦ז⸤⫴ְָ׆׸荈⹛⻉ׁ׸׷ךכ娖〷涸䗳搫

堣唒㷕统ַ׵堣唒涪鋅פ ˖ 堣唒涪鋅䪮遭ך灇瑔㛇湍ה׃גꬊ䌢חꅾ銲ⱄ植䚍٥㾩➂䚍זוر٦ةך颵הꆀך然⥂ /FHBUJWFر٦ة׾《׷㹋꿀װٓٝتي㹋꿀כر٦ة猰㷕♳כ䗳銲׌ָ➂꟦כװ׶׋ֻזְ˘ Science 363 (2019) Nature 583 (2020)
Nature Reviews Drug Discovery 17 (2018) 㹋꿀荈⹛⻉ך䪮遭涸涪㾜⻉㷕ד׮ꬊ⸬桦ז⸤⫴ְָ׆׸荈⹛⻉ׁ׸׷ךכ娖〷涸䗳搫

堣唒㷕统ַ׵堣唒涪鋅פ ˖ 堣唒涪鋅䪮遭ך灇瑔㛇湍ה׃גꬊ䌢חꅾ銲ⱄ植䚍٥㾩➂䚍זוر٦ةך颵הꆀך然⥂ /FHBUJWFر٦ة׾《׷㹋꿀װٓٝتي㹋꿀כر٦ة猰㷕♳כ䗳銲׌ָ➂꟦כװ׶׋ֻזְ˘ ˖ 㹋꿀荈⹛⻉ָ㹋植ׁ׸ג׮չ䌢חמהַֽ׵ך鿇ⴓ䞔㜠׃ַ䩛חⰅ׵זְպ劤颵כ㢌׻׵זְ Science 363 (2019) Nature
583 (2020) Nature Reviews Drug Discovery 17 (2018) 㹋꿀荈⹛⻉ך䪮遭涸涪㾜⻉㷕ד׮ꬊ⸬桦ז⸤⫴ְָ׆׸荈⹛⻉ׁ׸׷ךכ娖〷涸䗳搫

堣唒㷕统ַ׵堣唒涪鋅פ ˖ 堣唒涪鋅䪮遭ך灇瑔㛇湍ה׃גꬊ䌢חꅾ銲ⱄ植䚍٥㾩➂䚍זוر٦ةך颵הꆀך然⥂ /FHBUJWFر٦ة׾《׷㹋꿀װٓٝتي㹋꿀כر٦ة猰㷕♳כ䗳銲׌ָ➂꟦כװ׶׋ֻזְ˘ ˖ 㹋꿀荈⹛⻉ָ㹋植ׁ׸ג׮չ䌢חמהַֽ׵ך鿇ⴓ䞔㜠׃ַ䩛חⰅ׵זְպ劤颵כ㢌׻׵זְ ˖ 涪鋅ָ荈⹛⻉דֹ׷ַכ"*ⴓꅿחה׏ג׮琎䎃ך劢鍑寸㉏겗կչ➂䊨濼腉պ׾⡲׶׋ְז׵ 猘׋׍ָ傈ղ㼭ׁזչ涪鋅պהչ㷕统պ׾粸׶鵤׃ג⚅歲׾椚鍑׃גְֻ麓玎ך椚鍑כ♶〳鼘 Science
363 (2019) Nature 583 (2020) Nature Reviews Drug Discovery 17 (2018) 㹋꿀荈⹛⻉ך䪮遭涸涪㾜⻉㷕ד׮ꬊ⸬桦ז⸤⫴ְָ׆׸荈⹛⻉ׁ׸׷ךכ娖〷涸䗳搫

3FBMJUZ#JUFT麣Ⱗכ麩ִוծ׫׿זずׄ㉏겗ה䨌׏גְ׷˘ All models are wrong, but some are useful (George
Box) With enough data, the numbers speak for themselves. (Chris Anderson) If data can speak for themselves, they can also lie for themselves (David Hand) So it's critically important to - exercise caution - do not claim too much - understand the data - and its quality

ؚٖ؎نحؙأ剑黝⻉ ر٦ةず⻉邌植㷕统 ٌرٕك٦أ剑黝⻉ ٌرٕك٦أ䱱稊鷵如㹋꿀鎘歗 ًة㷕统垷⦺㷕统窟鎘涸锷椚さ䧭欰䧭涸ٌرٕ⻉
穗꿀锷堣唒㷕统 הさ椚锷椚锷٥锷椚䱿锷٥鎸〾Ⳣ椚 ך輐さ 5IFPSZESJWFO %BUBESJWFO ˖ 㼎韋植韋װ銲実ך醱꧟⻉椚锷⻉ד⢪ֲ⟎㹀ה㹋植韋ך鍑ꨄ ˖ ءىُٖ٦ءّٝ䪮岀ך醱꧟⻉䛬䠐䚍ך넝ְػًٓةװ鵚⡂䪮岀 ˖ 㼭؟ٝفٕ٥⡚ؕؐٝزך㉏겗 繎欰Ꟍ⸬卓ה6OEFSTQFDJDBUJPO ˖ 䌓秛غ؎،أךرؠ؎ٝ ˖ ✮庠ה椚鍑湱ꟼה㔓卓 ך彘 ˖ 穈さׇ涸䱱稊瑞꟦ך然桦涸䱱稊 ˖ 濼陎ك٦أה锷椚䱿锷鎸〾"* ךꣲ歲 ˖ ⿑㺘䱿锷װ䱱稊ך鎘皾旗涪 /1㔭ꨇ䚍 ˖ 㣐ꆀر٦ةך濼陎⻉ך㉏겗 ˖ ⵖ秈فؚٗٓىؚٝװ穈さׇ剑黝⻉ ➂䊨濼腉ⴓꅿ ➂䊨濼腉ⴓꅿ ˖ չ涪鋅ⶼ涪ⶼ鸡պװչ椚鍑պך锷椚 ˖ 僇爙涸濼陎ך栻䖤ך锷椚 ˖ ػة٦ٝ钠陎穗꿀ה⹞ ה锷椚涸䙼罋锷椚䱿锷 ךٍؘحف ˖ ٌرٕ黝欽眔㔲ה⥋걾䚍٥㸜Ⰻ䚍 սさ椚 ⯓꿀 锷վ ս穗꿀锷վ

⡭锑鎉铂זו麓玎ָ⯋ղ錁庠דֹזְ㼎韋׌הծ״׶겥衼 ➂䊨濼腉㷕⠓钞䎃䋆〾Q ؔ٦فٝ،ؙإأ https://doi.org/10.11517/jjsai.27.3_273

⡭锑剑鵚ך㹋⢽չ"MQIB'PMEWպ

⡭锑剑鵚ך㹋⢽չ堣唒㷕统ח1VSF.BUIװ׵ׇג׫׋պ

⡭锑剑鵚ך㹋⢽չ%.害ꟼ侧պ

自己紹介：機械学習・機械発見とデータ中心的自然科学

自己紹介：機械学習・機械発見とデータ中心的自然科学

More Decks by itakigawa

Other Decks in Research

Featured

Transcript