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
MolGANの紹介
Search
Oshita Noriaki
June 07, 2019
Education
0
1.2k
MolGANの紹介
Oshita Noriaki
June 07, 2019
Tweet
Share
More Decks by Oshita Noriaki
See All by Oshita Noriaki
第二回 3Dなんでも勉強会
ooshita
1
1.2k
Dockerで機械学習サービスの環境を作ろう!!
ooshita
0
230
事前準備
ooshita
0
200
Edge TPUで 発芽予測をする話
ooshita
0
2.3k
情報幾何の応用と最近の機械学習の動向
ooshita
2
4.1k
Other Decks in Education
See All in Education
Padlet opetuksessa
matleenalaakso
9
15k
XML and Related Technologies - Lecture 7 - Web Technologies (1019888BNR)
signer
PRO
0
3.2k
焦りと不安を、技術力に変える方法 - 新卒iOSエンジニアの失敗談と成長のフレームワーク
hypebeans
1
620
多様なメンター、多様な基準
yasulab
5
19k
ThingLink
matleenalaakso
28
4.2k
1216
cbtlibrary
0
120
Measuring your measuring
jonoalderson
0
210
あなたの言葉に力を与える、演繹的なアプローチ
logica0419
1
250
1202
cbtlibrary
0
170
HCI Research Methods - Lecture 7 - Human-Computer Interaction (1023841ANR)
signer
PRO
0
1.3k
Web Application Frameworks - Lecture 3 - Web Technologies (1019888BNR)
signer
PRO
0
3.2k
MySmartSTEAM 2526
cbtlibrary
0
170
Featured
See All Featured
Testing 201, or: Great Expectations
jmmastey
46
7.8k
How Fast Is Fast Enough? [PerfNow 2025]
tammyeverts
3
410
Sam Torres - BigQuery for SEOs
techseoconnect
PRO
0
150
The Hidden Cost of Media on the Web [PixelPalooza 2025]
tammyeverts
2
130
Building Applications with DynamoDB
mza
96
6.9k
[Rails World 2023 - Day 1 Closing Keynote] - The Magic of Rails
eileencodes
38
2.7k
JAMstack: Web Apps at Ludicrous Speed - All Things Open 2022
reverentgeek
1
300
Paper Plane (Part 1)
katiecoart
PRO
0
2.2k
Unlocking the hidden potential of vector embeddings in international SEO
frankvandijk
0
130
Evolving SEO for Evolving Search Engines
ryanjones
0
77
Making the Leap to Tech Lead
cromwellryan
135
9.7k
Agile Actions for Facilitating Distributed Teams - ADO2019
mkilby
0
97
Transcript
.PM("/ "OJNQMJDJUHFOFSBUJWFNPEFMGPSTNBMM NPMFDVMBSHSBQIT /JDPMB%F$BP 5IPNBT,JQG ൃදऀɿେԼൣߊ
ຊͷ༰ .PM("/ͱ (SBQI$POWPMVUJPOBM/FUXPSLͱ ("/ͱ .PM("/ͷΞʔΩςΫνϟʔ
ධՁࢦඪ ֶशʹ͓͚Δ
.PM("/ͱ w ͓͓ͬ͟ͺʹ͍͏ͱࢠάϥϑΛੜ͠Α͏ͱ͍͏("/
(SBQI$POWPMVUJPOBM/FUXPSL άϥϑϑʔϦΤม άϥϑԽ
ͳͥ($/Λ͏ͷʁ .4DIMJDIULSVMMFUBM ٙ 3//ϕʔεͷੜϞσϧͷจࣈྻදݱͰ͍͍Μ͡Όͳ͍ͷʁ ͳͥάϥϑʹʁ ͑ 3//ɺ ɾߏจنଇͱදݱͷॱংͷ͍͋·͍͞ͷ྆ํΛֶश͢ΔͨΊʹଟେͳίετΛඅ͢ ɾҰൠతʹʢඇࢠʣάϥϑʹద༻Ͱ͖ͳ͍
("/ min θ max ϕ x∼pdata (x) [logDϕ (x)] +
z∼pz (z) [log(1 − Dϕ (Gθ (z))] ผͰ͖ͳ͍Α͏ʹ͍ͨ͠ ผ͍ͨ͠ (ʹΑͬͯੜͨ͠σʔλ ຊͷσʔλ͔Βֶश
.PM("/ %JTDSJNJOBUPSσʔληοτͱ(FOFSBUPSΛࣝผ͢Δ 3FXBSEOFUXPSL(FOFSBUPSͰੜ͞ΕͨࢠΛධՁ͢Δɽ (FOFSBUPSࢠΛੜ͢Δɽ ɾͦΕͧΕͷ
matrixX = [x1 , …, xN] T ∈ ℝN×T ֤ࢠແάϥϑͰදݱͰ͖Δ
લఏɿ A ∈ ℝN×N×YXIFSFAij ∈ ℝY ྡςϯιϧ ऍߦྻ
3FXBSE/FUXPSL w άϥϑΈࠐΈʹجͮ͘ॱྻෆมหผث͓Αͼใु ωοτϫʔΫʢॴͷԽֶతੑ࣭ʹ͚ͨ3-ϕʔεͷ࠷ దԽͷͨΊʣ w ೖྗάϥϑɼग़ྗεΧϥʔ w %JTDSJNJOBUPSͱಉ͡ωοτϫʔΫΛ༻͢Δɽ
ධՁࢦඪ 4BNBOUBFUBM Uniqueness = ϢχʔΫαϯϓϧ ༗ޮαϯϓϧ Novelty = σʔληοτʹؚ·Εͳ͍༗ޮͳαϯϓϧ ༗ޮͳαϯϓϧ
Validity = 7BMJE શͯͷੜ͞Εͨࢠ
%SVHMJLFOFTTͲͷ͘Β͍Խ߹͕ༀʹͳΔՄೳੑ͕͋Δ͔ 4ZOUIFUJ[BCJMJUZࢠͷ߹ͷ༰қ͞ʢқʣ 4PMVCJMJUZࢠ͕Ͳͷఔਫੑ ਫʹ༹͚͍͔͢ Ͱ͋Δ͔
ࢠؒͷྨࣅͷܭࢉํ๏ʁ "#JDLFSUPOFUBMʹॻ͍͍ͯΔɽ 2&%ͱ͍͏ఆྔతͳਪఆΛߦ͍ͬͯΔɽdͷؒʹΛ࣋ͪɼ ʹ͍ۙ΄Ͳ%SVHMJLFOFTTΛͭɽ QED = exp ( 1 n
n ∑ i=1 ln di) ͜ͷத͕͚ۙΕༀΒ͍͠ͱ͍͑Δ EFTJSBCJMJUZGVODUJPOT B C D E F Gύϥϝʔλ d(x) = a + b [ 1 + exp (− x − c + d 2 e )] 1 − 1 [ 1 + exp (− x − c − d 2 f )] Yࢠهड़ࢠ
݁ 7"&ϕʔεͷੜϞσϧΑΓߴ͍༗ޮੑͱ৽نੑͷ྆ํͰ ࢠάϥϑΛੜ͢Δ͜ͱ͕Ͱ͖Δ ܭࢉ͕࣌ؒҎલͷγʔέϯγϟϧ("/ͱൺֱͯ͠ഒ͍܇࿅࣌ؒ ("/ͱڧԽֶशΛۦ͍ͯ͠ΔͷͰɼϞʔυ่յ͕ى͜Γ͍͢ ͦͷͨΊɼ৽حੑͷ͋Δάϥϑ͕ੜ͞Εʹ͘͘ͳΔ͋ Δɽ