$30 off During Our Annual Pro Sale. View Details »
Speaker Deck
Features
Speaker Deck
PRO
Sign in
Sign up for free
Search
Search
論文紹介:Relaxed Softmax for PU Learning
Search
Hayato Maki
October 05, 2019
Research
3
1k
論文紹介:Relaxed Softmax for PU Learning
論文リンク:
https://dl.acm.org/citation.cfm?id=3347034
Hayato Maki
October 05, 2019
Tweet
Share
More Decks by Hayato Maki
See All by Hayato Maki
Billion-scale Embedding for E-commerce Recommendation in Alibaba
hamaki
0
110
Today was a Good Day: The Daily Life of Software Developers
hamaki
0
110
MIRU 2019 Lunch on Seminar
hamaki
1
280
コーディネート整合性を考慮したカテゴリ間推薦
hamaki
0
1.2k
Regularization_The Element of Statical Learning
hamaki
0
190
Neural Activity During Sentence Processing as Reflected in Theta, Alpha, Beta, and Gamma Oscillations
hamaki
0
240
【ICML読み会】Unsupervised Deep Embedding for Clustering Analysis
hamaki
0
1.3k
Other Decks in Research
See All in Research
"主観で終わらせない"定性データ活用 ― プロダクトディスカバリーを加速させるインサイトマネジメント / Utilizing qualitative data that "doesn't end with subjectivity" - Insight management that accelerates product discovery
kaminashi
15
16k
大学見本市2025 JSTさきがけ事業セミナー「顔の見えないセンシング技術:多様なセンサにもとづく個人情報に配慮した人物状態推定」
miso2024
0
190
日本語新聞記事を用いた大規模言語モデルの暗記定量化 / LLMC2025
upura
0
370
Unsupervised Domain Adaptation Architecture Search with Self-Training for Land Cover Mapping
satai
3
390
その推薦システムの評価指標、ユーザーの感覚とズレてるかも
kuri8ive
1
280
Aurora Serverless からAurora Serverless v2への課題と知見を論文から読み解く/Understanding the challenges and insights of moving from Aurora Serverless to Aurora Serverless v2 from a paper
bootjp
1
140
[論文紹介] Intuitive Fine-Tuning
ryou0634
0
160
学習型データ構造:機械学習を内包する新しいデータ構造の設計と解析
matsui_528
4
2.1k
競合や要望に流されない─B2B SaaSでミニマム要件を決めるリアルな取り組み / Don't be swayed by competitors or requests - A real effort to determine minimum requirements for B2B SaaS
kaminashi
0
330
Nullspace MPC
mizuhoaoki
1
500
一人称視点映像解析の最先端(MIRU2025 チュートリアル)
takumayagi
6
4.4k
論文紹介:Not All Tokens Are What You Need for Pretraining
kosuken
1
220
Featured
See All Featured
Product Roadmaps are Hard
iamctodd
PRO
55
12k
Rails Girls Zürich Keynote
gr2m
95
14k
For a Future-Friendly Web
brad_frost
180
10k
Designing for Performance
lara
610
69k
Exploring the Power of Turbo Streams & Action Cable | RailsConf2023
kevinliebholz
37
6.2k
How to Ace a Technical Interview
jacobian
281
24k
Easily Structure & Communicate Ideas using Wireframe
afnizarnur
194
17k
Imperfection Machines: The Place of Print at Facebook
scottboms
269
13k
Building an army of robots
kneath
306
46k
Facilitating Awesome Meetings
lara
57
6.7k
Become a Pro
speakerdeck
PRO
31
5.7k
Optimising Largest Contentful Paint
csswizardry
37
3.5k
Transcript
3FMBYFE4PGUNBYGPS 16-FBSOJOH 6HP5BOJFMJBOBOE'MBWJBO7BTJMF $SJUFP3FTFBSDI 3FD4ZT`จհBU8BOUFEMZΦϑΟε հऀɿਅ༐ਓBU;0;0ςΫϊϩδʔζ
3FMBYFE4PGUNBYGPS 16-FBSOJOH 6HP5BOJFMJBOBOE'MBWJBO7BTJMF $SJUFP3FTFBSDI 3FD4ZT`จհBU8BOUFEMZΦϑΟε 1PTJUJWFBOE6OMBCFMFE հऀɿਅ༐ਓBU;0;0ςΫϊϩδʔζ
1PTJUJWF6OMBCFMFE 16 -FBSOJOH 3 ਖ਼ྫʹ͚ͩϥϕϧ͕͍͍ͭͯΔ ϥϕϧ͕͍͍ͭͯͳ͍σʔλɺਖ਼ྫɾෛྫͲͪΒͷ Մೳੑ͋Γ ਪનɺݴޠϞσϧͷֶशͰΑ͋͘Δঢ়گ
ී௨ͷ̎Ϋϥεྨ 16-FBSOJOH ਖ਼ྫ ෛྫ ਖ਼ྫ ϥϕϧແ͠
$POUFYUVBM16-FBSOJOH 4 ʮϖΞʯʹͳΔͷΛ୳͢16-FBSOJOH ਖ਼ྫͷϖΞ͚͕ͩ༩͑ΒΕΔ ਖ਼ྫͱͯ͠༩͑ΒΕͳ͔ͬͨϖΞͷΫϥεෆ໌ ֶश࣌ B@ C@
B@ C@ B@O C@O B@ C@ B@ C@ ϖΞ̌ ϖΞ ϖΞO ਖ਼ྫ Ϋϥεෆ໌ ਪ࣌ ೖྗ ީิ܈ ʁ ʁ ʁ ɾ ɾ ɾ ɾ ɾ ɾ ɾ ɾ ɾ
ϑΝογϣϯʹ͓͚Δ$POUFYUVBM16-FBSOJOH 5 ͷΈ߹ΘͤΛϞσϧԽ͢Δ ਖ਼ྫϖΞίʔσΟωʔτڞ༗αΠτ͔ΒೖखՄೳ ෛྫϖΞཧܥͷେֶʹߦͬͯࣸਅΛࡱΔखʹೖΒͳ͍ ʹߘ͞Εͨ τοϓεͱϘτϜεͷϖΞ Έସ͑ͨϖΞ
ؔ࿈ݚڀ 6 ਖ਼ྫͷੜ͚ͩΛֶश͢Δ ҟৗݕͳͲͰΘΕ͍ͯΔΞϓϩʔν ϥϯΩϯάͷUPQ/͚ͩʹڵຯ͕͋Δਪનʹ͔ͳ͍ ೳಈֶश ஞ࣍తʹϥϕϧΛ͚͍ͭͯ͘
0SBDMF ਅͷϥϕϧΛ͚ΒΕΔԿ ͔ ͕ඞཁɺਓؒͷΞϊςʔλʔͳͲ ϥϕϧ͖ σʔλ Ϟσϧֶश ϥϕϧແ͠ σʔλ 0SBDMF ೖྗ ద༻ αϯϓϦϯά ϥϕϧ༩
ຊݚڀͷߩݙ 7 $POUFYUVBM16-FBSOJOHʹ͍ͭͯɺ ৽͍͠ଛࣦؔͱෛྫαϯϓϦϯάͷ ํ๏ΛఏҊ
͖݅ͷ࠷ਪఆ 8 w J͕༩͑ΒΕͨ࣌ͷKͷ͖݅ܦݧ ਖ਼ྫͷϖΞ ਖ਼ྫϖΞҎ֎ͷશͯͷΈ߹Θͤ w ෛͷର GVMMTPGUNBY w
ϖΞͷू߹ɿ w JͱKͷϕΫτϧදݱͷੵɿ
͖݅ͷ࠷ਪఆ 9 w J͕༩͑ΒΕͨ࣌ͷKͷ͖݅ܦݧ ਖ਼ྫͷϖΞ ਖ਼ྫϖΞҎ֎ͷશͯͷΈ߹Θͤ w ෛͷର GVMMTPGUNBY
ͲͷJʹରͯ͠ ֤Kಉ͡ස ֬ Ͱෛྫͱͯ͠αϯϓϧ ͞ΕΔ ਅͷ͖݅ʹ JJEΛԾఆ w ϖΞͷू߹ɿ w JͱKͷϕΫτϧදݱͷੵɿ
ఏҊ
3FMBYFE4PGUNBY 34 -PTT 11 ෛྫαϯϓϦϯάɹɹʹ͍ͭͯɺଛࣦΛҎԼͷࣜͰఆٛ 'VMMTPGUNBY '4 ͱͷൺֱ
3FMBYFE4PGUNBY 34 -PTT 12 ෛྫαϯϓϦϯάɹɹʹ͍ͭͯɺଛࣦΛҎԼͷࣜͰఆٛ 'VMMTPGUNBY '4 ͱͷൺֱ
ͯ͢ͷKʹ͍ͭͯܭࢉ 2@J͕Ұ༷ͳΒɺ34'4ͷϞϯςΧϧϩۙࣅ 2@JͰαϯϓϧ͞ΕͨKͷू߹ 7 J K ʹ͍ͭͯܭࢉ
ੵΛߟྀͨ͠ෛྫαϯϓϦϯά 13 3FMBYFETPGUNBYΛඍ /VNCFSPGOFHBUJWFTBNQMFTˠ♾ ੵͷFYQΛ2@JͰॏΈ͚ ͨ͠ϘϧπϚϯ
ϘϧπϚϯෛྫαϯϓϦϯά 14 ෛྫʹͳΓ͢͞ʢOFHBUJWJUZʣͷࣄલࣝΛදݱ 5Λௐ͢Δ͜ͱͰෛྫαϯϓϦϯάͷੑ࣭Λௐ͢Δ ɿEFHFSBDZEJTUSJCVUJPO ॖୀ 5ɿԹύϥϝʔλ
ෛྫαϯϓϦϯάʹ͓͚ΔτϨʔυΦϑ 15 wϞσϧ͕ޡྨ͍͢͠ෛྫΛαϯϓϦϯά͍ͨ͠ ޡྨ͍͢͠ਖ਼ྫͱͷੵ͕େ͖͍ʢྨࣅ͕େ͖͍ʣ ຊਖ਼ྫͷσʔλΛɺෛྫͱͯ͠αϯϓϧͯ͠͠·͏ةݥ͕େ͖͍ wຊਖ਼ྫͷσʔλΛෛྫͱͯ͠αϯϓϦϯάͨ͘͠ͳ͍ ਖ਼ྫͱͷੵ͕খ͍͞ʢྨࣅ͕খ͍͞ʣσʔλΛαϯϓϦϯά
Ϟσϧʹͱͬͯ༰қʹࣝผՄೳɻใྔ͕গͳ͍ ϘϧπϚϯͷԹύϥϝλͰௐ
Թύϥϝλ 16 5ˠͷ࣌ɺਖ਼ྫͱͷੵ͕େ͖͍σʔλΛத৺ʹ୳ࡧ 5ˠ♾ͷ࣌ɺਖ਼ྫͱͷੵ͕খ͍͞σʔλΛத৺ʹ୳ࡧ 5 5 5
࣮ݧ
࣮ݧɿਓσʔλ ࠞ߹ਖ਼ن1@JͰσʔλΛੜ ਅͷͱਪఆ͞Εͨͷ,-μΠόʔδΣϯεͰධՁ ෛྫͷαϯϓϦϯάʹઃఆ ϘϧπϚϯͷઃఆ ܇࿅
ςετ ܇࿅ ςετ ܇࿅σʔλສɺόοναΠζΛมԽ όοναΠζɺ܇࿅σʔλΛมԽ 18
αϯϓϦϯά ଛࣦ .-& 'VMM 4PGUNBY 44 1PQVMBSJUZ 4PGUNBY 64 6OJGPSN
3FMBYFE4PGUNBY 6#4 #PUM[NBOO VOJGPSN 3FMBYFE4PGUNBY 14 1PQVMBSJUZ 3FMBYFE4PGUNBY 1#4 #PUM[NBOO QPQVMBSJUZ 3FMBYFE4PGUNBY #PUM[NBOOαϯϓϦϯάͷ༗ޮੑ͕֬ೝͰ͖Δ ࣮ݧɿਓσʔλ 19 όοναΠζ
࣮ݧɿ/FYUJUFN1SFEJDUJPO w .PWJFMFOT ສϨίʔυ w /FUqJYʢສϨίʔυʣ w 8JLJQFEJB5FYU ࠷ϙϐϡϥʔͳສ୯ޠ
w ̏୯ޠҎʹొͨ͠ΒϙδςΟϒͱΈͳ͢ 20 σʔλ ධՁࢦඪ w w .13 .FBO1FSDFOUJMF3BOL w 1SFDJTJPOBU
݁Ռɿ/FYUJUFN1SFEJDUJPO 21 αϯϓϦϯά ଛࣦ .-& 'VMM 4PGUNBY 44 1PQVMBSJUZ 4PGUNBY
64 6OJGPSN 3FMBYFE4PGUNBY 6#4 #PUM[NBOO VOJGPSN 3FMBYFE4PGUNBY 14 1PQVMBSJUZ 3FMBYFE4PGUNBY 1#4 #PUM[NBOO QPQVMBSJUZ 3FMBYFE4PGUNBY
·ͱΊ 22 $POUFYUVBM16-FBSOJOHʹ͍ͭͯɺ3FMBYFE4PGUNBYଛࣦͱ ϘϧπϚϯෛྫαϯϓϦϯάΛఏҊ ਓσʔλɺ/FYUJUFNQSFEJDUJPOͰ༗ޮੑΛ֬ೝ Թύϥϝʔλͷಈతௐ
$//ɺ3//ͳͲʹର͢ΔదԠ ܭࢉྔ͕ଟͦ͏ʁ จλΠτϧͷ͚͔͕͍ͭͨͬͨͳ͍ 'VUVSF8PSL ॴײ