Upgrade to Pro
— share decks privately, control downloads, hide ads and more …
Speaker Deck
Features
Speaker Deck
PRO
Sign in
Sign up for free
Search
Search
マイクロソフト発の麻雀AIの論文「Suphx: Mastering Mahjong with ...
Search
lyakaap
February 25, 2025
0
2
マイクロソフト発の麻雀AIの論文「Suphx: Mastering Mahjong with Deep Reinforcement Learning」 について
マイクロソフト発の麻雀AIの論文「Suphx: Mastering Mahjong with Deep Reinforcement Learning」 についての解説
lyakaap
February 25, 2025
Tweet
Share
More Decks by lyakaap
See All by lyakaap
大規模画像テキストデータのフィルタリング手法の紹介
lyakaap
7
1.6k
Multi-view 3D Human Pose Estimationの最近の動向
lyakaap
2
3.8k
Shopee - 2nd Place Solutionと上位解法まとめ
lyakaap
0
1.1k
Kaggle Carvanaコンペでの取り組み
lyakaap
1
6k
Featured
See All Featured
Cheating the UX When There Is Nothing More to Optimize - PixelPioneers
stephaniewalter
280
13k
実際に使うSQLの書き方 徹底解説 / pgcon21j-tutorial
soudai
175
52k
A designer walks into a library…
pauljervisheath
205
24k
Facilitating Awesome Meetings
lara
52
6.2k
jQuery: Nuts, Bolts and Bling
dougneiner
63
7.6k
Art, The Web, and Tiny UX
lynnandtonic
298
20k
XXLCSS - How to scale CSS and keep your sanity
sugarenia
248
1.3M
Navigating Team Friction
lara
183
15k
Testing 201, or: Great Expectations
jmmastey
42
7.2k
Imperfection Machines: The Place of Print at Facebook
scottboms
267
13k
Large-scale JavaScript Application Architecture
addyosmani
511
110k
Let's Do A Bunch of Simple Stuff to Make Websites Faster
chriscoyier
507
140k
Transcript
剑䓼ך띾ꦾ"*չ4VQIYպך鍑铡 埆㽵 ⥜䎂 !MZBLBBQ
כׄח • 4VQIY.BTUFSJOH.BIKPOHXJUI%FFQ3FJOGPSDFNFOU-FBSOJOH IUUQTBSYJWPSHBCT 铣ךד鍑铡׃תׅ • و؎ؙٗاؿزך灇瑔ث٦يח״锷俑 • طحز띾ꦾ٦ي㣓뒉ד띾ꦾ"*ⴱך⼧媮麦䧭♳⡘ •
띾ꦾךٕ٦ٕח鑫׃ֻזֻג椚鍑דֹ״ֲחծ䪮遭涸ז銲稆ח ꅾֹ縧ְ鍑铡חזַה䙼תׅ
䲿
띾ꦾחאְג • ⴓ겲♳כչ㢳➂侧♶㸣Ⰻ䞔㜠٦يպ • فٖ؎َ٦ַ鋅ִגְזְ䞔㜠ָ֮٦ي • ه٦ؕ٦ֿ • ٕ٦ָٕ醱꧟ •
麊ך銲稆ַָז窃 • 䓼ְ䓲ְך鐰⣣ָ㣐㢌
띾ꦾךٕ٦ٕ馄ֻׂ • 㔊➂ד挿侧㤽ְさֲ • ٦ي穄✪儗ך䭯挿ָ넝ְ➂קו♳ך갫⡘ח • 醱侧ךٓؐٝس㽷הㄎלָ֮ • ٓؐٝس׀הח挿侧ך㟓幾ָ涪欰
• 窩さׇػؤٕ♳䩛חװה⸇挿 https://tenhou.net/
4VQIYך嚊銲
䩛岀ך銲秈 • 醱侧ך殴鴥صُ٦ٕٓطحزٙ٦ؙ$//欽ְ • 㷕统ך崧 ➂ךفٖ؎ر٦ة⢪ג侄䌌֮㷕统 ̔荈䊹㼎䨌ח״䓼⻉㷕统דׁח䓼ֻ •
ئַٗ䓼⻉㷕统ׅ״㷕统ָ㸜㹀 • 䓼⻉㷕统ח띾ꦾךسً؎ٝ暴䚍ח䘔ׄ䊨㣗⸇ִ
4VQIYך⚥魦
4VQIYך⚥魦 • 띾ꦾך⚺銲ז،ؙءّٝ䰍ג晤鼅䫛٥هٝ٥ث٦٥ؕٝ٥ٔ٦ث • ך،ؙءّٝⴻ倖䬐䔲ׅ杝甧׃$//ٌرٕא • ٌرٕ圓鸡כקר♧筰Ⰵ⸂ך暴䗙ꆀָ蕯䎁殯ז • Ⰵ⸂湍ך朐䡾 •
⳿⸂չ֮،ؙءّٝ《ַやַպ 䰍ג晤 هٝ ث٦ ؕٝ ٔ٦ث 湍 ؒٝ؝٦س Ⰵ⸂ 4VQIY ⳿⸂
ٌرٕךⰅ⸂暴䗙ꆀ⡲䧭 • 띾ꦾדכչ♲٥㔊٥✿պך״ֲח鸬竲׃侧㶵ך晤䲧ִה㴍׃ְ • 晤ך侧㶵ך鸬竲ꟼ⤘ָ䠐䭯א ̔䩛䭯ך晤זוך湍ך䞔㜠♴㔳ך״ֲחؒٝ؝٦س 䩛䭯ך晤 ♧卐⟃♳䨽䭯 ✳卐⟃♳䨽䭯
♲卐⟃♳䨽䭯 㔊卐⟃♳䨽䭯
ٌرٕךⰅ⸂暴䗙ꆀ⡲䧭 䩛䭯ך晤 ♧卐⟃♳䨽䭯 ✳卐⟃♳䨽䭯 ♲卐⟃♳䨽䭯 㔊卐⟃♳䨽䭯 ך殴鴥 • ٌرٕך鎘皾儗חכ♴㔳ך״ֲח殴鴥ת˟
˟ז簚倯ぢ罋䣁׃殴鴥ח׃זַךַכ♶僇
➭ך暴䗙ꆀ • 뒊ַגְ晤 • سٓ • 䰍ג晤 • 䭯挿 •
• MPPLBIFBEGFBUVSF如ل٦آד铡僇 ˟ 暴䗙ꆀך侧כⰋ鿇ד_⦐玎䏝
暴䗙ꆀ-PPLBIFBEGFBUVSF • ֮䩛䭯ך晤❛䳔׃הֹח וךְּך挿侧וךְּך然桦ד栻䖤דֲַֹ • ⢽匌䰍ג㜥さծ挿ծ挿ד،ֲؖ • 2וֲװג鎘皾ׅך • "ֲ֮،ؖٔךػة٦ٝ帾ׁ⮚⯓䱱稊ד実
• 㼎䨌湱䩛ך遤⹛כ搀鋔׃גծ荈ⴓך䩛䭯ך晤ֽ罋䣁 ،ؖ䩛㸣䧭ׇֿׁהד挿侧栻䖤ׅ遤捀
4VQIYךٌرٕ圓鸡 䰍ג晤鼅䫛ٌرٕ هٝ٥ث٦٥ؕٝ٥ٔ٦ثٌرٕ https://dmv.nico/ja/articles/mahjong_ai_naga/ PS
㷕统،ٕ؞ٔؤي
㷕统ך崧 ➂ךفٖ؎ر٦ة⢪ג侄䌌֮㷕统 • 暴ח㢌ֿהכ׃גְזְךדⶴ䠥 ̔荈䊹㼎䨌ח״䓼⻉㷕统דׁח䓼ֻ • %JTUSJCVUFE3FJOGPSDFNFOU-FBSOJOH •
1PMJDZHSBEJFOU • *NQPSUBODFTBNQMJOH • &OUSPQZSFHVMBSJ[BUJPO • (MPCBM3FXBSE1SFEJDUJPO • 0SBDMF(VJEJOH • 1BSBNFUSJD.POUF$BSMP1PMJDZ"EBQUBUJPOֿ搀׃ד⼧媮זג
• %JTUSJCVUFE3FJOGPSDFNFOU-FBSOJOH • 1PMJDZHSBEJFOU • *NQPSUBODFTBNQMJOH • &OUSPQZSFHVMBSJ[BUJPO • (MPCBM3FXBSE1SFEJDUJPO
• 0SBDMF(VJEJOH
%JTUSJCVUFE3FJOGPSDFNFOU-FBSOJOHⴓ侔ءأذيך嚊銲 • 4FMGQMBZXPSLFST • ٦يךءىُٖ٦ة٦$16؟٦غ٦ • 䱿锷ؒٝآٝ(16؟٦غ٦ • 1BSBNFUFSTFSWFS •
剑倜ך倯瘻ٌرٕךػًٓ٦ة • 麓ךٔفٖ؎غحؿ؋
%JTUSJCVUFE3FJOGPSDFNFOU-FBSOJOH倯瘻ך栻䖤 • 倯瘻⺟ꂁ岀ك٦أ • ꅾ挿؟ٝفؚٔٝהؒٝزٗؾ٦姻⻉遤ֲֿהד㷕统⤛鹌 麓ח֮ת䱱稊׃גְזְהֿ ⮚⯓׃ג䱱稊ׅ״ֲחׅ ٌثك٦ءّٝ倯瘻חְְ䠬ׄךغٓ ٝأךٓٝتي䚍䭯ְׇ
ؒٝزٗؾ٦ָ㼭ֺׁׅ ̔寸ת遤⹛׃ַ《זֻזךד㽷 䨽鍑חכתג׃תֲ ؒٝزٗؾ٦ָ㣐ֹֺׅ ̔⽃秪ח㷕统ָ㸜㹀׃זְ ꅾ挿؟ٝفؚٔٝ ؒٝزٗؾ٦姻⻉
%JTUSJCVUFE3FJOGPSDFNFOU-FBSOJOHؔؿٓ؎ٝ鐰⣣ • 䓼⻉㷕统ח״גչ侄䌌֮㷕统ךպךהֹ״䓼ֻז • ˟㸜㹀媮⡘䓼ׁ邌ׅ䭷垥דծ넝ְקו䓼ְ 侄䌌֮㷕统ך 㸜㹀媮⡘ 䓼⻉㷕统
• %JTUSJCVUFE3FJOGPSDFNFOU-FBSOJOH • 1PMJDZHSBEJFOU • *NQPSUBODFTBNQMJOH • &OUSPQZSFHVMBSJ[BUJPO • (MPCBM3FXBSE1SFEJDUJPO
• 0SBDMF(VJEJOH
(MPCBM3FXBSE1SFEJDUJPOٌثك٦ءّٝ • 띾ꦾךٕ٦ٕ_ٓؐٝس㽷嶊⻉䖓ך剑穄أ؝،ד갫⡘寸㹀 • 㣓뒉ך㜥さծ갫⡘ח䘔ׄג㜠ꂹه؎ٝزָ䖤 • ⢽♧⡘ QU ✳⡘ QU
♲⡘QU 㔊⡘QU • 剑穄갫⡘ח״㜠ꂹؿ؍٦سغحؙה׃ג⢪欽׃㜥さ ♳䩛חفٖ؎דֹٓؐٝسהֲדזְٓؐٝس⼒ⴽדֹזְ • ٓؐٝس׀הךأ؝،㟓幾ؿ؍٦سغحؙה׃㜥さ أ؝،幾ׅչ葺ְ،ؙءّٝպָ㶷㖈ׅ˟姻䔲ח،ؙءّٝך葺 ׃䝤׃鐰⣣דֹזְ ̔葺ְ䠬ׄךؿ؍٦سغحָؙדֹזְַ ˟زحف然㹀ׇׁח剑穄ٓؐٝسד佦䠐ח䮶鴥 זו
(MPCBM3FXBSE1SFEJDUJPO • ٓؐٝس׀הח٦يך剑穄涸ז㜠ꂹ鋅琎✮庠ٌرٕ㼪Ⰵ HMPCBMSFXBSEQSFEJDUPS • ٌرٕכ✳㾴ך(36ה.-1ח״ג圓䧭 • 㹋ꥷךفٖ؎ر٦ةהח㷕统 üֿח״ٓؐٝس׀הך㜠ꂹ黝ⴖח鋅琎ֿהָ〳腉 •
植㖈ךٓؐٝسהٓؐٝسך✮庠⦼ך䊴ծ植㖈ךٓؐٝسך 㜠ꂹה׃ג欽ְ
(MPCBM3FXBSE1SFEJDUJPOٌرٕ圓鸡 剑ⴱךٓؐٝسַ植㖈ךٓؐٝسתדך䞔㜠ך禸
(MPCBM3FXBSE1SFEJDUJPOؔؿٓ؎ٝ鐰⣣ • 㷕统חׯה顀柃׃גְֿהָⴓַ 侄䌌֮㷕统ך 㸜㹀媮⡘ 䓼⻉㷕统 䓼⻉㷕统 (MPCBM3FXBSE1SFEJDUJPO
• %JTUSJCVUFE3FJOGPSDFNFOU-FBSOJOH • 1PMJDZHSBEJFOU • *NQPSUBODFTBNQMJOH • &OUSPQZSFHVMBSJ[BUJPO • (MPCBM3FXBSE1SFEJDUJPO
• 0SBDMF(VJEJOH
0SBDMF(VJEJOH • 띾ꦾדכفٖ؎َ٦ָ鋅ֿהָדֹזְ䞔㜠ְָםְ֮ • ָֿؒ٦آؑٝزך㷕统ꨇ׃ֻ׃גְ銲㔓 • ֿד㷕统أؾ٦س鸞חծⰋגך䞔㜠鋅ֿהָ דֹչؙٕؔٓպד㷕统酡⸔ • ♧➂ֽ㸣Ⰻ䞔㜠٦ي朐䡾
• דכؙٕؔٓוֲװגⵃ欽ׅךַ
0SBDMF(VJEJOHؙٕؔٓךⵃ欽倯岀 • ַؙٕؔٓ鋅ִגְ䞔㜠㷕统ך鹌遤ח䘔ׄג媮ղהꦀ׃גְֻ • 剑穄涸חכ鸐䌢ךؒ٦آؑٝزהずׄ朐䡾חז • 襑殅כ♳䩛ֻ遤ַזַ׃ְ Ⰻ鿇鋅ִ 鸐䌢הずׄ
㷕统ך穗麓
0SBDMF(VJEJOHؔؿٓ؎ٝ鐰⣣ • 䗍䊴דכ֮ךך㷕统חׯה顀柃׃גְ 侄䌌֮㷕统ך 㸜㹀媮⡘ 䓼⻉㷕统 䓼⻉㷕统 (MPCBM3FXBSE1SFEJDUJPO 䓼⻉㷕统
(MPCBM3FXBSE1SFEJDUJPO 0SBDMF(VJEJOH
ؔٝٓ؎ٝ鐰⣣
ؔٝٓ؎ٝ鐰⣣ • 㹋ꥷך䚍腉鐰⣣ׅחؔٝٓ؎ٝ띾ꦾ؟؎زך㣓뒉⢪ג 鐰⣣遤 • 穠卓ה׃גכ傀㶷ך띾ꦾ"*ֶ״ן➂ךزحففٖ؎َ٦㣐ֹֻ ♳㔐䚍腉爙׃
ؔٝٓ؎ٝ鐰⣣ • ׁח⼧媮חⵋ麦ծ➂فٖ؎َ٦ך״♳ך㹋⸂爙׃
תה
תה • و؎ؙٗاؿز涪ך띾ꦾ"*չ4VQIYպָ㣓뒉⼧媮麦䧭 • 圫ղז䊨㣗ד㷕统ך㸜㹀⻉װ䚍腉ぢ♳ • 侄䌌֮㷕统̔䓼⻉㷕统 • 䓼⻉㷕统ח띾ꦾךسً؎ٝ暴䚍ח䘔ׄ䊨㣗⸇ִ