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
20191005_Recsys2019_Users_In_The_Loop.pdf
Search
Waku Michishita
October 05, 2019
Research
260
0
Share
20191005_Recsys2019_Users_In_The_Loop.pdf
Waku Michishita
October 05, 2019
More Decks by Waku Michishita
See All by Waku Michishita
Rでグラフ作るの!
wmichi
0
750
Bayesian Optimizationを用いたパラメータチューニング
wmichi
0
790
数式と実装から復習するツリーアルゴリズム
wmichi
0
390
Other Decks in Research
See All in Research
ペットのかわいい瞬間を撮影する オートシャッターAIアプリへの スマートラベリングの適用
mssmkmr
0
460
Using our influence and power for patient safety
helenbevan
0
330
2026-01-30-MandSL-textbook-jp-cos-lod
yegusa
1
1.1k
[チュートリアル] 電波マップ構築入門 :研究動向と課題設定の勘所
k_sato
0
400
[BlackHatAsia2026] Hidden Telemetry: Uncovering TraceLogging ETW Providers You're Not Using (Yet)
asuna_jp
1
330
通時的な類似度行列に基づく単語の意味変化の分析
rudorudo11
0
260
ScoreMatchingRiesz for Automatic Debiased Machine Learning and Policy Path Estimation with an Application to Japanese Monetary Policy Evaluation
masakat0
0
260
Tiaccoon: Unified Access Control with Multiple Transports in Container Networks
hiroyaonoe
0
1.6k
さくらインターネット研究所テックトーク2026春、研究開発Gr.25年度成果26年度方針
kikuzo
0
130
AI Agentの精度改善に見るML開発との共通点 / commonalities in accuracy improvements in agentic era
shimacos
6
1.6k
「行ける・行けない表」による地域公共交通の性能評価
bansousha
0
140
ブレグマン距離最小化に基づくリース表現量推定:バイアス除去学習の統一理論
masakat0
0
240
Featured
See All Featured
Avoiding the “Bad Training, Faster” Trap in the Age of AI
tmiket
0
130
Designing for Performance
lara
611
70k
The AI Revolution Will Not Be Monopolized: How open-source beats economies of scale, even for LLMs
inesmontani
PRO
3
3.4k
Building Flexible Design Systems
yeseniaperezcruz
330
40k
The Cost Of JavaScript in 2023
addyosmani
55
9.9k
How to Ace a Technical Interview
jacobian
281
24k
Exploring the Power of Turbo Streams & Action Cable | RailsConf2023
kevinliebholz
37
6.4k
JavaScript: Past, Present, and Future - NDC Porto 2020
reverentgeek
52
5.9k
Kristin Tynski - Automating Marketing Tasks With AI
techseoconnect
PRO
0
230
Jamie Indigo - Trashchat’s Guide to Black Boxes: Technical SEO Tactics for LLMs
techseoconnect
PRO
0
120
Unlocking the hidden potential of vector embeddings in international SEO
frankvandijk
0
770
Dominate Local Search Results - an insider guide to GBP, reviews, and Local SEO
greggifford
PRO
0
150
Transcript
6TFSTJOUIF-PPQ "1TZDIPMPHJDBMMZ*OGPSNFE"QQSPBDIUP 4JNJMBS*UFN3FUSJFWBM ͷհ 3FD4ZTจಡΈձ!8BOUFEMZ ಓԼ ٱ
ຊհ͢Δจ • λΠτϧ 6TFSTJOUIF-PPQ"1TZDIPMPHJDBMMZ*OGPSNFE "QQSPBDIUP4JNJMBS*UFN3FUSJFWBM • ஶऀ "NZ"8JOFDPGG 'MPSJO#SBTPWFBOV #SZDF
$BTBWBOU 1FBSDF8BTIBCBVHI BOE.BUUIFX (SBIBN 5SVF'JU
֓ཁ • ఏҊ .BUIFNBUJDBMͳྨࣅͰͳ͘1TZDIPMPHJDBMMZ *OGPSNFEͳྨࣅΛϨίϝϯυʹ༻͍ͯͲ͏͔ʁ • ҙࣝ ΦϑϥΠϯධՁͱΦϯϥΠϯධՁཱ͕྆͠ͳ͍ .BUIFNBUJDBMͳྨࣅ͕Ϣʔβͷ৺ཧతཁҼΛߟྀ ͍ͯ͠ͳ͍͔ΒͰʁ
͜ͷจͰΓ͍ͨ͜ͱ 1TZDIPMPHJDBMMZ*OGPSNFEͳྨࣅͷଌఆ ํ๏ͷཱ֬ 1TZDIPMPHJDBMMZ*OGPSNFEͳྨࣅͷఆࣜԽ
)VNBO1FSDFQUJPOBOE+VEHFNFOUT ਓؒͷ֮Λௐࠪ͢Δํ๏৭ʑ͋Δ • Α͘༻͍ΒΕΔͷ 4JOHMF*UFN"QQSPBDI • ࠓճ͍͍ͨͷ 'PSDFE$IPJDF
4JOHMF*UFN"QQSPBDI ҎԼͷΑ͏ͳ࣭Λ͛ͯධՁͤ͞Δํ๏
4JOHMF*UFN"QQSPBDIͷܽ 4JOHMF*UFN"QQSPBDIͷલఏͷ͍͔ͭ͘ ࣮ࡍʹΓཱͨͳ͍ͱࢦఠ • UIFTDBMFJTVTFEJOUIFTBNF VOCJBTFE XBZCZBMMQBSUJDJQBOUT ˡ࣮ࡍྑ͍ํʹධՁ͕ͪͩͬͨ͠Γ͢Δ • IVNBOTIBWFTUBCMF
BCTPMVUFQSFGFSFODF ˡ࣮ࡍจ຺ೝͷڐ༰ྔʹΑ੍ͬͯݶ͞ΕΔ
'PSDFE$IPJDF ҎԼͷΑ͏ʹೋऀҰͰ͑ͤ͞Δܗࣜ ͋ͳ͕ͨམͱͨ͠ͷ ۚͷ佁ʁͦΕͱۜͷ佁ʁ
'PSDFE$IPJDFͷརͱܽ • ར • ईͷղऍʹᐆດ͕͞ͳ͍ • Ϣʔβ͕Ͳ͜Ͱࣅ͍ͯΔʗࣅ͍ͯͳ͍ͷڥքઢΛҾ ͍ͨͷ͔ཧղ͍͢͠ • ܽ
• ϢʔβʹΑͬͯڥքઢΛͻ͘5ISFTIPME͕ҟͳΔ • ࣮ࡍʮͲͪΒબͳ͍ʯͱ͍͏બ͕͋Γ͏Δ ͨͩ͠ࠓճٻΊ͍ͯΔճʮͲͪΒ͕ൺֱతࣅ͍ͯΔ ͔ʯͳͷͰ͜ͷҰ୴ແࢹ͢Δ͜ͱͱ͍ͯ͠Δɻ
4JNJMBSJUZͷධՁ ैདྷͷํ๏ Ϩίϝϯσʔγϣϯʹ͓͍ͯΑ͘༻͍ΒΕΔͷ NFUSJDEJTUBODFGVODUJPOϕʔεͷͷ FY +BDDBSE 4JNJMBSJUZ " ,
= | ∩ | | ∪ | • ه߸ͷఆٛͦΕͧΕҎԼͷ௨Γ , +BDDBSEʹجͮ͘ྨࣅ , ྨࣅΛௐΔΞΠςϜ , ֤ΞΠςϜͷಛྔ
ܽ ैདྷͷࢦඪͰҎԼͷ͜ͱ͕ߟྀͰ͖͍ͯͳ͍ͱࢦఠ • ྨࣅੑఆͷඇରশੑ ʮ"ͱ#ࣅ͍ͯΔ͔ʁʯͱʮ"#ʹࣅ͍ͯΔ͔ʁʯ Ͱ݁Ռ͕ҟͳΓ͏Δ • ಛྔͷྨࣅͷد༩ͷҧ͍ ࣮ࡍϢʔβ͕֮͢Δಛྔͷد༩ҟͳΔͷͰʁ
4JNJMBSJUZͷධՁ ఏҊ͞Εͨํ๏ • 5WFSTLZNPEFM • ಛྔ͝ͱͷॏཁΛߟྀͨ͠ྨࣅ ҎԼͷΑ͏ʹఆࣜԽ͞ΕΔ - ,
= ∩ − − − ( − ) • ه߸ͷఆٛ ∩ ͱͷ྆ํʹؚ·ΕΔಛྔ − ʹͷΈଘࡏ͢Δಛྔ − ʹͷΈଘࡏ͢Δಛྔ , , ্هͷͦΕͧΕͷॏཁΛௐ͢Δύϥϝʔλ
࣮ݧํ๏ UBSHFUʢਅΜதʣʹ͍ͭͯͲͪΒͷBMUFSOBUJWF͕ ΑΓࣅ͍ͯΔͷ͔͑ͤ͞Δ ࣮ݧͷࡉ͔͍ઃఆจࢀর
ྨࣅͷఆࣜԽ • NVMUJMFWFM-PHJTUJD3FHSFTTJPOΛ༻͍ͯ • ͦΕͧΕͷಛྔͷد༩ͷҧ͍Λ໌Β͔ʹ͢Δ • +BDDBSEͱͷྨࣅͷҧ͍Λ໌Β͔ʹ͢Δ • 5WFSTLZNPEFMͷྨࣅ -5,6
= - -∩: − - -;: − (:;-) = (, , ) ֤ಛྔͷॏΈ ͱॏཁͷύϥϝʔλ -∩: ྆ऀʹڞ௨ͷಛྔ͔Ͳ͏͔Λࣔ͢\ ^ͷϕΫτϧ -;: λʔήοτ5ʹͷΈʹଘࡏ͢Δಛྔ͔Ͳ͏͔Λࣔ͢ :;- ൺֱର"ʹͷΈʹଘࡏ͢Δಛྔ͔Ͳ͏͔Λࣔ͢
༻ͨ͠ಛྔ 4-NBUDI4MFFWF-FOHUI͕Ұக͢Δ͔ %- NBUDI%SFTT-FOHUI͕Ұக͢Δ͔ 14 NBUDI1BUUFSO4UZMF͕Ұக͢Δ͔ $' NBUDI$PMPS'BNJMZ͕Ұக͢Δ͔ %4 NBUDI%SFTT4IBQF͕Ұக͢Δ͔
/- NBUDI/FDL-JOF͕Ұக͢Δ͔ %' NBUDI%SFTT'JU͕Ұக͢Δ͔
ଛࣦؔͱ࠷దԽ • ଛࣦؔ = − 1 ? @AB C @
log @Δ- + 1 − @ log 1 − @ ΔSL ℎ @- = B BPQRS ;TU5 V , Δ- = -5,6 − -5,W () • ࠷దԽ minV − B [ log − B [ log − B [ log ℎ ≫ 0 ℎ −
݁Ռ֓ཁ ಛྔ͝ͱʹد༩͕ҧ͏ -PHJTUJD3FHSFTTJPOͷ݁Ռ5WFSTLZ NPEFMͷํ͕ྑ͔ͬͨ ʢ͔֓͠͠Ͷࣅͨ݁Ռʹͳ͍ͬͯΔʣ Ұ෦Ͱྨࣅʹ͕ࠩݱΕͨ
ଐੑ͝ͱͷد༩ͷҧ͍ • Ұ୴୯७ͳ-PHJTUJD3FHSFTTJPOͰಛྔ͝ͱ ͷӨڹΛ֬ೝ • ಛྔ͝ͱʹد༩ҟͳΔʢ5BCMFʣ ͜ͷ࣌Ͱ·ͩ5WFSTLZNPEFMΛ͍ͬͯͳ͍
ଐੑ͝ͱͷد༩ͷҧ͍ • 5WFSTLZNPEFMͰઌͷ݁Ռͱಉ༷ಛྔ͝ͱ ʹد༩͕ҧ͏ • , ʹ͕ࠩग़͓ͯΓఆʹඇରশੑ͕͋Δ ͱΘ͔Δ ্هGPMEͷฏۉΛͱͬͨͷ
݁Ռ ྨࣅੑఆͷҧ͍ • ϩδεςΟοΫͷ݁Ռ5WFSTLZNPEFMͷํ ͕Α͔ͬͨ • จͰݴٴͷΈ • ͔ͨͩ͠ͳΓࣅͨΑ͏ͳ݁Ռʹͳ͍ͬͯͨΒ͍͠
• ྨࣅʹ͕ࠩݱΕͨ • ྨࣅͷఆ͕͍͠Έ߹Θͤ ΄Ͳ྆ऀʹ͕ࠩͰ͍ͯΔ • Ұํ͔ͳΓྨࣅ͍ͯ͠Δ PSશ͘ྨࣅͯ͠ͳ͍߹ͷ ͔ࠩͳΓখ͍͞
՝ • ʮબ͠ͳ͍ʯͱ͍͏બࢶ ͜ͷબࢶ͕͋Δ߹৺ཧతͳಈ͖ʹӨڹ͕͋Γ͏Δ • ϨίϝϯυͰͲ͜·Ͱ͑Δ͔ະ ࠓճղ͍ͨͷ͋͘·Ͱྨ