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
[計算機構論] Why do tree-based models still outperfo...
Search
mei28
December 06, 2022
0
43
[計算機構論] Why do tree-based models still outperform deep learning on tabular data?
計算機構論の発表資料
Why do tree-based models still outperform deep learning on tabular data?(NeurIPS2022)
mei28
December 06, 2022
Tweet
Share
More Decks by mei28
See All by mei28
[読み会] “Are You Really Sure?” Understanding the Effects of Human Self-Confidence Calibration in AI-Assisted Decision Making
mei28
0
49
[JSAI'24] 人間の判断根拠は文脈によって異なるのか?〜信頼されるXAIに向けた人間の判断根拠理解〜
mei28
1
390
[CHI'24] Fair Machine Guidance to Enhance Fair Decision Making in Biased People
mei28
0
34
[DEIM2024] 卓球の得点予測における重要要素の分析
mei28
0
25
[Human-AI Decision Making勉強会] 意思決定 with AIは個人vsグループで変わるの?
mei28
0
170
[読み会] Words are All You Need? Language as an Approximation for Human Similality Judgements
mei28
0
28
[参加報告] AAAI'23
mei28
0
78
[計算機構論] Learning Models of Individual Behavior in Chess
mei28
0
66
チーム開発と機械学習
mei28
0
44
Featured
See All Featured
Optimising Largest Contentful Paint
csswizardry
33
2.9k
Performance Is Good for Brains [We Love Speed 2024]
tammyeverts
3
360
Typedesign – Prime Four
hannesfritz
39
2.4k
Bash Introduction
62gerente
608
210k
Cheating the UX When There Is Nothing More to Optimize - PixelPioneers
stephaniewalter
280
13k
[RailsConf 2023 Opening Keynote] The Magic of Rails
eileencodes
28
9k
Unsuck your backbone
ammeep
668
57k
Six Lessons from altMBA
skipperchong
26
3.4k
The Illustrated Children's Guide to Kubernetes
chrisshort
48
48k
Art, The Web, and Tiny UX
lynnandtonic
296
20k
Happy Clients
brianwarren
97
6.7k
Optimizing for Happiness
mojombo
376
69k
Transcript
ܭࢉػߏ! ༶໌ 8IZEPUSFFCBTFENPEFMTTUJMM PVUQFSGPSNEFFQMFBSOJOHPOUBCVMBSEBUB
ࣗݾհ • ໊લɿ༶໌ • ॴଐɿ૯߹จԽݚڀՊҬՊֶઐ߈അݚڀࣨ% ʢ݄ೖֶͰஜେ͔Βདྷ·ͨ͠ʣ • ݚڀ༰ɿਓؒͱػց͕ڠௐͰ͖ΔΑ͏ͳػցֶश
ɹɹɹɹˠम࢜Ͱɼҙࢥܾఆͷެฏੑ͕ςʔϚ • ͻͱ͜ͱɿྑ͍ͯͩ͘͘͠͞ʂ
จใ • ϑϥϯεͷํʑɼ*OSJB4BDMBZύϦαΫϨʔେֶͱ͔ ͱ࿈ܞ͍ͯ͠Δݚڀػؔ • બΜͩཧ༝ • ,BHHMFͳͲͰʮॳख-JHIU(#.ʯ͕ओྲྀͰ͋Γɼ.-1 ͕ڧ͍ͱݶΒͳ͍ཧ༝ΛΓ͔͔ͨͬͨΒɽ
͜ͷจ͔ΒಘΒΕΔͷ ͳͥςʔϒϧσʔλͰɼਂϞσϧܾఆΛ ྇կͰ͖ͳ͍ͷ͔ ˠ࣮ݧΛ௨ͯ͠ཧ༝Λߟ͍ͯ͘͠ ςʔϒϧσʔλͷͨ͘͞ΜͷϕϯνϚʔΫΛ࡞ • ༻ͨ͠σʔληοτެ։
• ͞ΒʹɼϋΠύϥௐࡁΈͷϞσϧެ։
༻͢Δ༻ޠʹ͍ͭͯ • ܾఆɿ෦ͷΞϧΰϦζϜ͕ܾఆͳͷશͯΛ͢͞ • //Tɿଟύʔηϓτϩϯਂχϡʔϥϧωοτϫʔ ΫͳͲɼܾఆͰͳ͍ΞϧΰϦζϜΛࢦ͢ɽ
എܠɿςʔϒϧσʔλʹܾఆ͕ओྲྀ ը૾ɼࣗવݴޠɼԻͳͲͷͰɼ ਂֶशϞσϧͷੑೳ͕΄ͱΜͲ4P5"Λͱ͍ͬͯΔ ҰํͰɼςʔϒϧσʔλͰ͍·ͩʹܾఆ͕ڧ͍ //Tσʔλʹରͯ͠ɼؼೲόΠΞεΛڧ͘ੜΉ͕͋ Δ
• ؼೲόΠΞεɿϞσϧʹΑΔԿΒ͔ͷ੍ʢFHઢܗʣ
എܠɿදܗࣜʹಛԽͨ͠//Tͳ͍Θ͚Ͱͳ͍ ςʔϒϧσʔλΛରͱͨ͠//Tݚڀ͞Ε͍ͯΔ ఏҊख๏ʹΑͬͯɼैདྷͷܾఆΑΓ ༏ҐͰ͋Δओுଟ͘ଘࡏ • ಛఆͷσʔληοτʢఏҊऀ࡞ͳͲʣʹͷΈ༏Ґ •
ͨ·ͨ·ϋΠύϥ͕Α͔ͬͨͷͰ ධՁํ๏͕ΒΒͳͷͰɼൺֱͰ͖ͳ͔ͬͨ ˠຊจͰɼϕϯνϚʔΫΛ࡞͢Δʂ
എܠɿܾఆ͕ͳͥڧ͍ͷ͔ //T͕ͳͥऑ͍ͷ͔ //T͕ςʔϒϧσʔλʹ͓͍ͯؼೲόΠΞεʹΑͬͯɼ ੑೳ্͕͍ͯ͠ͳ͍ͷܦݧతʹΒΕ͍ͯΔ ͔͠͠ɼͲΜͳཁૉ͕ؼೲόΠΞεΛͨΒ͍ͯ͠Δ͔ ɼಥ͖ࢭΊ͍ͯΔͷͳ͍ɽ ˠຊจʹΑͬͯ͜ͷݪҼʹ͍ͭͯಥ͖ࢭΊΔ
ຊݚڀͷߩݙ ςʔϒϧσʔλΛ༻͍ͨϕϯνϚʔΫΛ࡞ɽ • ϋΠύϥௐͨ͠Ϟσϧར༻ՄೳͳܗͰެ։ //TͱܾఆΛطଘͷσʔληοτΛ༻͍ͯൺֱ ͍ΖΜͳ݅ઃఆɼϋΠύϥௐΛߦ͏ɽ
ςʔϒϧσʔλͰͳܾͥఆ͕//TΑΓ ڧ͍ͷ͔Λಥ͖ࢭΊͨ
ϕϯνϚʔΫσʔληοτΛ࡞Δ • ϞσϧͷൺֱΛߦ͏ͨΊʹɼϕϯνϚʔΫΛ࡞Δɽ • ࠓճ͞·͟·ͳυϝΠϯͷςʔϒϧσʔλΛछྨ࡞ ͢Δɽ •
Ͳͷσʔληοτ0QFO.-Ͱఏڙ͞Ε͍ͯΔɽ • ࣍ͷϖʔδͷ͜ͱʹҙͯ͠બΜͩ
σʔληοτબʢʣ )FUFSPHFOFPVTDPMVNO • ֤ྻੑ࣭ͱͯ͠ҟͳΔը૾ηϯαʔใෆՄ /PUIJHI%JNFOTJPOBM • σʔληοτͷରͯ͠ߴ࣍ݩͰͳ͍࠷ߴͰ࣍ݩ
6OEPDVNFOUFEEBUBTFUT • ใྔҙຯ͕Θ͔Βͳ͍σʔλΛΘͳ͍ ҉߸Խ͞Ε͍ͯΔ
σʔληοτબʢʣ **%EBUB • ࣌ܥྻσʔλͳͲɼ**%Ͱͳ͍σʔλΘͳ͍ 3FBMXPSMEEBUB • γϛϡϨʔγϣϯਓσʔλ༻͍ͳ͍ɽ
/PUUPPTNBMM • ಛྔ͕ҎԼ αϯϓϧ͕ҎԼΘͳ͍
σʔληοτબʢʣ /PUUPPFBTZ • ༧ଌ͕؆୯ͳσʔληοτΘͳ͍ • σϑΥϧτͷϩδεςΟοΫճؼͱ3FT/FUͷείΞ͕ ૬ରͰҎͩͬͨΒ࠾༻͠ͳ͍ɽ /PUEFUFSNJOJTUJD
• ༧ଌ͕σʔλʹରܾͯ͠ఆతʹܾ·Δͷ༻͍ͳ͍ • ϙʔΧʔνΣεͷήʔϜใͳͲ
෭࣍తͳղফ͢Δʢʣ ςʔϒϧσʔλΛֶश͢ΔλεΫͱͯ͠Γ͚ΔͨΊʹ ࣍ͷʹؾΛ͚ͭΔ .FEJVNTJ[FEUSBJOJOHTFU • αϯϓϧʹͳΔΑ͏ʹσʔλΛΔ
/PNJTTJOHEBUB • ܽଛશͯআɽ
෭࣍తͳղফ͢Δʢʣ #BMBODFEDMBTTFT • ྨλεΫʹ͓͍ͯɼϥϕϧൺಉఔʹ -PXDBSEJOBMJUZDBUFHPSJDBMGFBUVSF • ྻͷதʹछྨҎ্ͷΧςΰϦมআ )JHIDBSEJOBMJUZOVNFSJDBMGFBUVSF •
ྻͷதʹछྨҎԼͷมআ • ͔ͭ͠ͳ͍࣌ΧςΰϦมͱͯ͠ѻ͏
ϋΠύϥௐ ϋΠύϥௐɼϞσϧੑೳ্ʹͪΐͬͱߩݙ͢Δ ϋΠύϥௐʹؔͯ͠)ZQFSPQUΛͬͨ ϥϯμϜαʔνΛ͢Δ ֤σʔληοτͰΠςϨʔγϣϯͰύλʔϯߦ͍ɼྑ ͔ͬͨͷΛ࠷ྑϋΠύϥͱ͢Δ ࣌ؒతʹ ίϯϐϡʔλ࣌ؒͷ݁ՌΛެ։ͨ͠
ධՁࢦඪ ྨͰਫ਼ɼճؼͰ3είΞΛධՁࢦඪʹ͢Δ ҟͳΔσʔληοτͰείΞΛൺֱ͢Δͷࠔ • ճؼͩͱɼΒ͖ͭ߹͍͕͜ͱͳΔͨΊ ˠ"WFSBHFEJTUBODFUPUIFNJOJNVNΛ࠾༻͢Δ • ࠷ྑͱ࠷ѱͷείΞΛͱʹਖ਼نԽ͢Δ
σʔλͷલॲཧ ೖྗʹ༻͍ΔࡍͷલॲཧΛҰ؏ͯ͠ߦ͏ (BVTTJBOJ[FEGFBUVSF • มਖ਼نʹج͍ͮͯ࠶ม 5SBOTGPSNFESFHSFTTJPOUBSHFUT • తมΛมֶ͠शɽਪ࣌ٯม͢͠ 0OF)PU&ODPEFS
• ΧςΰϦมϫϯϗοτԽ͢Δ
ର߅͢ΔܾఆͷҰཡ ʢจதʹॻ͔Ε͍ͯͳ͔ͬͨͷͰਪͰ͢ʣ 9(#PPTUɿͨͿΜ(16ʹରԠ͍ͯ͠Δ͔Β 3BOEPN'PSFTUɿݹయత͕ͩڧ͍ (SBEJFOU#PPTUJOH5SFFɿΧςΰϦมѻ͑Δ
ର߅͢Δ//TͷҰཡ .-1ɿ3FEVDF0O1MBUFBVͷεέδϡʔϥΛՃ 3FT/FUɿ.-1 ESPQPVU CBUDIMBZFSOPSNBMJ[F TLJQDPOOFDUJPO '5@5SBOTGPSNFSɿมΛຒΊࠐΊΔUSBOTGPSNFSϞσϧ
ςʔϒϧσʔλͰҰ൪༗ྗ 4"*/5ɿ5SBOTGPSNFS JOUFSTBNQMFTBUUFOUJPOMBZFS ɹɹɹɹPVUQFSGPSN͢Δ͔Β࠾༻
݁ՌɿมͷΈͷͱ͖ • υοτઢσϑΥϧτ • ϋΠύϥௐͯ͠//T4P5"ʹͳΒͳ͍ • ࣮ߦ࣌ؒॻ͍ͯͳ͍͕ɼܾఆͷํ͕ૣ͘ऴΘΔ
݁Ռɿม ΧςΰϦม • طଘݚڀͰදܗࣜͷΧςΰϦมͷѻ͍͕૪Ͱ͋ͬͨ • ΧςΰϦมͷ͍ͤͰ//T͕ऑ͍Θ͚Ͱͳ͍
ͳͥ//T͕ܾఆʹউͯͳ͍͔Λߟ͍͑ͯ͘ ϕϯνϚʔΫ࣮ݧʹΑΓɼ//T͕ܾఆʹউͯͳ͍͜ͱΛ ࠶֬ೝͨ͠ ͔͜͜ΒɼςʔϒϧσʔλͷಛྔΛมԽͤ͞ɼҧ͍ʹ ͍ͭͯ୳ڀ͍ͯ͘͠ ඪɿදܗࣜͷͲ͏͍͏ಛ͕ܾఆʹ༗ޮͰ //TͰ͍ͯ͘͠͠Δཁૉͳͷ͔Λݟ͚ͭΔ
//Tͷग़ྗΒ͔ʹͳΔΑ͏ʹภΔ ֤܇࿅ͷηοτͷग़ྗʹରͯ͠ɼΧʔωϧฏԽΛߦ͏ɽ • Χʔωϧฏ͔Λߦ͏ͱɼ֎Εͷ༧ଌΛܰݮ͢Δ͜ͱ ͕Մೳɽ • ΧʔωϧฏԽۙ͘ͷͷͷ͍ۙΠϝʔδ
//Tͷग़ྗΒ͔ʹͳΔΑ͏ʹภΔ • ฏԽͷ෯Λมߋ͢Δͱɼ ܾఆͷਫ਼͕ஶ͘͠མͪ ͨ • ҰํͰ//T͕ࠩখ͍͞ ˠͱͱ//T͕ࣗΒ͔ ͳग़ྗΛߦ͏ؼೲόΠΞε͕
ଘࡏ
//Tͷग़ྗΒ͔ʹͳΔΑ͏ʹภΔ
//Tෆඞཁͳใʹऑ͍ • ϥϯμϜϑΥϨετʹΑͬͯϥϯΫ͚ͨ͠ಛྔॏཁ ʹԠͯ͡ಛྔΛܽམͨ࣌͠ͷੑೳࠩʹ͍ͭͯݟ͍ͯ͘ • (#%5ͰɼಛྔΛܽམͤͯ͞ɼਫ਼Լ͋· Γى͖ͳ͔ͬͨ • ܽམͤͨ͞ಛྔͷΈͰֶशͯ͠ਫ਼͋·Γ্͕Βͳ
͍ ˠ͖ͪΜͱඞཁͳใͷΈΛֶͬͯश͍ͯ͠Δ
//Tෆඞཁͳใʹऑ͍ • ಛྔͷܽམͱɼඇใͳಛྔΛՃͨ࣌͠ͷൺֱ • //Tඇใಛྔʹରͯ͠ؤ݈Ͱͳ͍
දܗࣜʹճసෆมੑ͕ͳ͘ɼ//Tͷֶशʹ͔ͳ͍ • //TճసෆมੑͷಛΛ࣋ͭɽͭ·ΓҙͷϢχλϦ ߦྻΛೖྗʹ͔͚ͯग़ྗʹӨڹ͕গͳ͍ɽ ճసෆมੑɿೖྗ͕ճసͯ͠ɼຊ࣭มΘΒͳ͍ ✖ ϢχλϦߦྻͷྫ
දܗࣜʹճసෆมੑ͕ͳ͘ɼ//Tͷֶशʹ͔ͳ͍ • σʔληοτΛϥϯμϜʹճసͤ͞ɼਫ਼ͷมԽΛݟΔ
·ͱΊ • ςʔϒϧσʔλʹ͓͍ܾͯఆͱ//Tͷҧ͍ΛΈ͚ͭͨ //Tͷ༧ଌΒ͔ʹͳΔؼೲόΠΞε //Tෆඞཁͳใʹऑ͍ දܗࣜͷճసෆมੑ͕ͳ͍͜ͱ͕//TͱϛεϚον • දܗࣜͷσʔληοτͱϋΠύϥௐࡁΈϞσϧΛެ։ ײɿ
• ৽ख๏ΛఏҊͨ͠Θ͚Ͱͳ͍͕ɼ࣮ݧͱߟ͕ஸೡ • ͓ۚͱ࣌ؒΛͨ͘͞Μඅ͍ͯ͠Δͷ͕͏͔͕͑ͨ
ϒϥβૢ࡞͢ΔͷʹϚε͍ͬͯΒͳ͘ͳ͍ʁʁ $ISPNFͷ֦ுػೳʮ7JNJVNʯΛ͓͢͢Ί͠·͢ʂʂ ϚεͷҠಈڑLN͋ΔΈ͍ͨ<>ɹ • ຖ͜Μͳಈ͔͢ͷ͠ΜͲ͍ • ΩʔϘʔυͱϚεͷԟ෮ແବ͡Όͳ͍ʁ ͣͬͱΩʔϘʔυ্ʹखΛஔ͍ͯ࡞ۀͰ͖ͨΒ
ΜΓͳͷʹͳ͊ <>IUUQTEBJMZQPSUBM[KQLJKJNPVTF@QPJOUFSNPWFNFOU@EJTUBODF
ϒϥβૢ࡞͢ΔͷʹϚε͍ͬͯΒͳ͘ͳ͍ʁʁ ͦΕɼʮWJNJVNʯͰͰ͖·͢ ϒϥδϯάͷϚεૢ࡞Λ ΩʔϘʔυͷΩʔૢ࡞Ͱସ͢Δ͜ͱ͕Մೳʂ • ૢ࡞ײWJNʹ͍ۙͨΊɼWJNNFS͙͢׳ΕΔʂ ݸਓతʹWJNJVNͷΦϜχݕࡧ͕ΊͬͪΌศར IUUQTDISPNFHPPHMFDPNXFCTUPSFEFUBJMWJNJVNECFQHHFPHCBJCIHOIIOEPKQFQJJIDNFC