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
ファッションのコーディネートを自動生成してみた/FashionTech Talks Tokyo...
Search
tn1031
May 14, 2016
2
1.1k
ファッションのコーディネートを自動生成してみた/FashionTech Talks Tokyo #1 LT
tn1031
May 14, 2016
Tweet
Share
More Decks by tn1031
See All by tn1031
Outfit Generation and Style Extraction via Bidirectional LSTM and Autoencoder
tn1031
0
120
インタラクティブな属性操作が可能なファッションアイテム検索/attribute manipulation survey
tn1031
0
1.1k
Autoencoderを用いたOutfitからのスタイル抽出/style auto encoder
tn1031
0
12k
fashion_workshop_survey/Size Recommendation System for Fashion E-commerce
tn1031
0
280
画像を用いたファッションアイテム検索/Image Retrieval for Fashion
tn1031
0
5.5k
ファッションアイテム検索における深層学習の活用/Fashion Item Retrieval using Deep Learning
tn1031
0
2.3k
ディープラーニングでコーデを提案/FashionTechMeetup#4
tn1031
0
2.3k
KDD 2016勉強会/Images Don’t Lie: Transferring Deep Visual Semantic Features to Large-Scale Multimodal Learning to Rank
tn1031
0
1k
Fashion Tech x Machine Learning/twm_fashion_ml
tn1031
5
5.6k
Featured
See All Featured
ピンチをチャンスに:未来をつくるプロダクトロードマップ #pmconf2020
aki_iinuma
124
52k
KATA
mclloyd
29
14k
How to train your dragon (web standard)
notwaldorf
92
6.1k
The Language of Interfaces
destraynor
158
25k
Writing Fast Ruby
sferik
628
61k
Making Projects Easy
brettharned
116
6.3k
A better future with KSS
kneath
239
17k
Become a Pro
speakerdeck
PRO
28
5.4k
What's in a price? How to price your products and services
michaelherold
245
12k
Measuring & Analyzing Core Web Vitals
bluesmoon
7
490
Mobile First: as difficult as doing things right
swwweet
223
9.7k
The Power of CSS Pseudo Elements
geoffreycrofte
77
5.8k
Transcript
ϑΝογϣϯͷίʔσΟωʔτ Λࣗಈੜͯ͠Έͨ 2016/5/13 FashionTech Talks Tokyo #1 @tn1031
ࣗݾհ ▸ தଜ ຏ / @tn1031 ▸ σʔλαΠΤϯςΟετ ▸ SIer(2)
-> VASILY ▸ ػցֶशΛઐ߈ ▸ ϊϯύϥϝτϦοΫϕΠζ ▸ มϕΠζ ▸ MCMC, SMC ▸ FashionؔͷSlide ▸ https://speakerdeck.com/tn1031/fashion-tech-meetup-number-2-lt ▸ https://speakerdeck.com/tn1031/twm-fashion-ml 2 @tn1031
VASILYͱiQONʹ͍ͭͯ
VASILYͱiQONʹ͍ͭͯ ϑΝογϣϯΞϓϦʮiQONʯΛӡӦ͍ͯ͠·͢ 4
VASILYͱiQONʹ͍ͭͯ Ϣʔβ͕ϑΝογϣϯΞΠςϜ ΛΈ߹ΘͤͯίʔσΛ࡞ΕΔ 5
VASILYͱiQONʹ͍ͭͯ ؾʹೖͬͨΞΠςϜͦͷͰECαΠτʹඈΜͰ͓ങ͍Ͱ͖Δ 6 ఏܞECαΠτ
VASILYͱiQONʹ͍ͭͯ ຊதͷECαΠτͷใΛΫϩʔϦϯά 7
VASILYͱiQONʹ͍ͭͯ iQON͕ѻ͏σʔλͷྫɿίʔσΟωʔτ 8 ίʔσ ‣ ΘΕͨΞΠςϜ ‣ ΞΠςϜͷϒϥϯυ ‣ ϨΠΞτ
‣ λΠτϧ ‣ ίϝϯτ ‣ λά ‣ ɾɾɾ
ίʔσΟωʔτΛ࡞ͬͯΈΔ
ίʔσΟωʔτΛ࡞ͬͯΈΔ iOSɿhttps://itunes.apple.com/jp/app/fasshonkodineto-iqon-aikon/id497264307 Androidɿhttps://play.google.com/store/apps/details?id=jp.vasily.iqon&hl=ja ϒϥβɿhttps://www.iqon.jp/edit/ 10 ΞϓϦ ϒϥβ
ίʔσΟωʔτΛ࡞ͬͯΈΔ 1. ΞΠςϜΛબͯ͠ 2. ஔ 11 ΞϓϦ ϒϥβ ᶃ ɹ
ᶄ ɹ ᶃ ɹ ᶄ ɹ
ίʔσΟωʔτΛ࡞ͬͯΈΔ 1. ΞΠςϜΛબͯ͠ 2. ஔ 12 ΞϓϦ ϒϥβ ͜Ε͚ͩ
ίʔσΟωʔτΛ࡞ͬͯΈΔ 1. ΞΠςϜΛબͯ͠ 2. ஔ 13 ΞϓϦ ϒϥβ ͜Ε͚ͩ ͳΜͰ͕͢ɺ
ίʔσΟωʔτΛ࡞ͬͯΈΔ 14 ϫϯϐʔε 15ΞΠςϜ x 335ϖʔδ εΧʔτ 15ΞΠςϜ x 256ϖʔδ
ύϯϓε 15ΞΠςϜ x 1108ϖʔδ ɾ ɾ ɾ
ίʔσΟωʔτΛ࡞ͬͯΈΔ 15 ‣ ΧςΰϦɿ ɹ70ΧςΰϦ ‣ 1ϖʔδ͋ͨΓͷΞΠςϜɿ ɹɹ15ΞΠςϜ ‣ ֤ΧςΰϦͷϖʔδɿ
ɹेʙઍ(!?)
ίʔσΟωʔτΛ࡞ͬͯΈΔ 16 ΞΠςϜଟ͗͢ɽɽɽ
ίʔσΟωʔτΛ࡞ͬͯΈΔ 17 ίʔσΟωʔτΛ ͬͱָʹ࡞Γ͍ͨ ࣗಈੜ͠Α͏ʂ
ίʔσΟωʔτΛࣗಈੜͯ͠ΈΔ
ίʔσࣗಈੜγεςϜͷ࡞Γํ 1. ΞΠςϜwordɺίʔσΟωʔτdocument 19 (w1, w2, . . . ,
wn) (I1, I2, . . . , In) ίʔσΟωʔτΛจॻͱΈͳ͠ɺɹɹɹɹɹɹ ςΩετϚΠχϯάͷख๏Λద༻͢Δ จॻ ୯ޠྻ ΞΠςϜྻ ίʔσΟωʔτ
ίʔσࣗಈੜγεςϜͷ࡞Γํ 2. RNNʹಥͬࠐΉ 20 http://qiita.com/moji_ai/items/87afd4a433dc655d8cfd inputɿΞΠςϜIDྻ outputɿ࣍ͷΞΠςϜͷ༧ଌ֬ RNNςΩετ࣌ܥྻσʔλͷղੳʹ Α͘༻͍ΒΕΔχϡʔϥϧωοτϫʔΫ outputɿ࣍ͷΞΠςϜͷ༧ଌ֬
ίʔσࣗಈੜγεςϜͷ࡞Γํ 3. ֶश͕͏·͍͘͘Α͏ʹفΔ 21 iteration perplexity
ίʔσࣗಈੜγεςϜͷ࡞Γํ 22 Ͱ͖·ͨ͠ɻ
ࣗಈੜͷ݁Ռ
ࣗಈੜΞϧΰϦζϜͰબΕͨΞΠςϜ 24
ࣗಈੜΞϧΰϦζϜͰબΕͨΞΠςϜ 25
ࣗಈੜΞϧΰϦζϜͰબΕͨΞΠςϜ 26
ॴײ·ͱΊͳͲ
ॴײ 28 ‣ Կʹ͏ͷʁ ‣ ίʔσੜ࣌ͷਪનγεςϜ(༧ଌมͷΠϝʔδ) ‣ ͳΜͰRNNʁ ‣ ΞΠςϜྻ͔͠Θͳ͍ͳΒɺͨͩͷ͖݅֬ΛٻΊΔ
ͳͷͰɺτϐοΫϞσϧͰྑͦ͞͏ ‣ ը૾Λ͍͍ͨ ‣ ίʔσΟωʔτݟ͕ͨେࣄ ‣ ը૾Λѻ͑ΔΑ͏ʹNNΛ֦ு͢Δඞཁ͕͋Δ
·ͱΊ ‣ RNNΛ༻͍ͯίʔσΟωʔτΛࣗಈੜͨ͠ 29 ‣ ը૾ͦͷଞͷଐใΛͬͯϞσϧΛߴԽ ‣ ΞΠςϜͷஔ·ͰࣗಈԽ ‣ ίʔσΟωʔτͷςΠετͷ੍ޚ
‣ ϢʔβʹύʔιφϥΠζͨ͠ਪનγεςϜ ·ͱΊ Future Works
͝ਗ਼ௌ͋Γ͕ͱ͏͍͟͝·ͨ͠ We are hiring !! ڵຯͷ͋ΔํͷೖࣾɾΠϯλʔϯΛ͓͓ͪͯ͠Γ·͢ʂʂ http://vasily.jp/