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
Google TensorFlowとAndroidが繋がる未来
Search
ARIYAMA Keiji
March 12, 2016
Technology
4
6.2k
Google TensorFlowとAndroidが繋がる未来
Android Bazaar and Conference 2016 Spring 発表資料
http://abc.android-group.jp/2016s/
ARIYAMA Keiji
March 12, 2016
Tweet
Share
More Decks by ARIYAMA Keiji
See All by ARIYAMA Keiji
Build with AI
keiji
0
220
DroidKaigi 2023
keiji
0
1.9k
TechFeed Conference 2022
keiji
0
290
Android Bazaar and Conference Diverse 2021 Winter
keiji
0
890
ci-cd-conference-2021
keiji
1
1.2k
Android Bazaar and Conference 2021 Spring
keiji
3
820
TFUG KANSAI 20190928
keiji
0
130
Softpia Japan Seminar 20190724
keiji
1
180
pixiv App Night 20190611
keiji
1
600
Other Decks in Technology
See All in Technology
ABEJA FIRST GUIDE for Software Engineers
abeja
0
3.2k
社内外から"使ってもらえる"データ基盤を支えるアーキテクチャの秘訣/登壇資料(飯塚 大地・高橋 一貴)
hacobu
PRO
0
8.3k
マルチドライブアーキテクチャ: 複数の駆動力でプロダクトを前進させる
knih
0
12k
Symfony AI in Action
el_stoffel
1
150
AI時代のインシデント対応 〜時代を切り抜ける、組織アーキテクチャ〜
jacopen
4
170
IaC を使いたくないけどポリシー管理をどうにかしたい
kazzpapa3
1
180
TypeScript 6.0で非推奨化されるオプションたち
uhyo
15
5.5k
Data Hubグループ 紹介資料
sansan33
PRO
0
2.3k
All About Sansan – for New Global Engineers
sansan33
PRO
1
1.3k
生成AIシステムとAIエージェントに関する性能や安全性の評価
shibuiwilliam
2
250
Eight Engineering Unit 紹介資料
sansan33
PRO
0
5.6k
Pandocでmd→pptx便利すぎワロタwww
meow_noisy
2
1k
Featured
See All Featured
Code Review Best Practice
trishagee
72
19k
VelocityConf: Rendering Performance Case Studies
addyosmani
333
24k
Designing for humans not robots
tammielis
254
26k
GraphQLとの向き合い方2022年版
quramy
49
14k
The Language of Interfaces
destraynor
162
25k
Navigating Team Friction
lara
190
16k
Fight the Zombie Pattern Library - RWD Summit 2016
marcelosomers
234
17k
The World Runs on Bad Software
bkeepers
PRO
72
12k
The MySQL Ecosystem @ GitHub 2015
samlambert
251
13k
What’s in a name? Adding method to the madness
productmarketing
PRO
24
3.8k
Thoughts on Productivity
jonyablonski
73
4.9k
The Success of Rails: Ensuring Growth for the Next 100 Years
eileencodes
46
7.8k
Transcript
C-LIS CO., LTD.
C-LIS CO., LTD. ༗ࢁܓೋʢ,FJKJ"3*:"."ʣ C-LIS CO., LTD. AndroidΞϓϦ։ൃऀ ػցֶशॳ৺ऀ
ͬͯ·ͤΜ Photo : Koji MORIGUCHI (AUN CREATIVE FIRM)
C-LIS CO., LTD.
C-LIS CO., LTD.
C-LIS CO., LTD. IUUQXXXNFCJDDPNDSFBUPSTpMFIUNM ϝϏοΫઔொɹΫϦΤΠλʔζϑΝΠϧ Vol.376
C-LIS CO., LTD. IUUQXXXNFCJDDPNDSFBUPSTpMFIUNM ϝϏοΫઔொɹΫϦΤΠλʔζϑΝΠϧ Vol.376
C-LIS CO., LTD.
C-LIS CO., LTD. ࿘ͷϏϧɾήΠπ ͳʹΘ IUUQXXXNFCJDDPNDSFBUPSTpMFIUNM ϝϏοΫઔொɹΫϦΤΠλʔζϑΝΠϧ Vol.376
(PPHMF5FOTPS'MPXͱ "OESPJE͕ܨ͕Δະདྷ
(PPHMF*0
C-LIS CO., LTD. 5FOTPS'MPXൃදʢʣ TM
ษڧձΖ͏ͥ
C-LIS CO., LTD. (%(,PCFʢBUେࡕۀେֶʣ
C-LIS CO., LTD. (%(,ZPUPʢBUۙـେֶʣ
C-LIS CO., LTD. ษڧձͰͷൃදςʔϚ IUUQTHEHLPCFEPPSLFFQFSKQFWFOUT
4PGUXBSF%FTJHO݄߸ʰ"OESPJE4UVEJPͷελΠϧͰޮΞοϓʱΑΓ © ࠜઇΕ͍ ؟ ڸ ͬ ່
Έͷ؟ڸ່ͬը૾ΛࣗಈͰऩू͍ͨ͠
C-LIS CO., LTD. ൃද༰ (%(,PCFʢBUେࡕۀେֶʣ
C-LIS CO., LTD. ൃද༰ (%(,ZPUPʢBUۙـେֶʣ
C-LIS CO., LTD. ࣮ݱΛ્Ή̏ͭͷน σʔλͷน ܭࢉྔͷน ػցֶशͷน
C-LIS CO., LTD.
C-LIS CO., LTD. σʔλͷน σʔλͷน େྔͷ܇࿅σʔλΛ୭͕Ͳ͏ͬͯ༻ҙ͢Δʁ ܭࢉྔͷน ػցֶशͷน
C-LIS CO., LTD. େྔͷσʔλΛ Ͳ͏ͬͯखʹೖΕΔ͔ copyright © 2016 Live2D
Inc. all rights reserved.
؟ڸ່ͬͷϞσϧΛ࡞Δʹ ؟ڸΛ֎ͨ͠ঢ়ଶͷֆ͕ඞཁ
ϓϩδΣΫτɺ ҉ʹΓ্͛ͨʜʜ
C-LIS CO., LTD. σʔλՃͷޮΞοϓ 0QFO$7ΛͬͯإΛΓग़͢ಛྔͷϑΝΠϧ͕ ެ։͞Ε͍ͯͨ 0QFO$7ʹΑΔΞχϝإݕग़ͳΒMCQDBTDBEF@BOJNFGBDF IUUQVMUSBJTUIBUFOBCMPHDPNFOUSZ
C-LIS CO., LTD. σʔλ࡞ ख࣋ͪͷ؟ڸ່ͬը૾ ຕΛॲཧ ຕͷը૾͔Β ຕͷإը૾ʢީิʣΛऔಘ ຕʹ͍ͭͯ0QFO$7͕إΛࣝผͰ͖ͣ
C-LIS CO., LTD. ࡞σʔλ༁ ຕதɺ؟ڸ່ͬը૾͕ຕ ͦΕҎ֎͕ຕ إը૾Ͱͳ͍ޡݕग़͕ຕ ؟ڸ່ͬ
ඇ؟ڸ່ͬ ޡݕग़
C-LIS CO., LTD. ܭࢉྔͷน σʔλͷน ܭࢉྔͷน खॻ͖จࣈೝࣝʢ./*45ʣͰ$16ͩͱͳ͔ͳ͔ऴΘΒͳ͍ ػցֶशͷน
C-LIS CO., LTD. (16ϚγϯΛಋೖ Graphic Memory: 2 GB compute
capability: 5.2 (F'PSDF(59
C-LIS CO., LTD. ͷൺֱ .BDʢ$16ʣ 2016-03-06 00:45:41.023059: step 0,
loss = 11.16 (37.3 examples/sec; 3.436 sec/batch) 2016-03-06 00:46:16.764922: step 10, loss = 11.05 (41.2 examples/sec; 3.108 sec/batch) 2016-03-06 00:46:55.245639: step 20, loss = 10.95 (37.4 examples/sec; 3.424 sec/batch) (16 2016-03-06 00:42:54.454566: step 90, loss = 10.39 (286.9 examples/sec; 0.446 sec/batch) 2016-03-06 00:42:59.533350: step 100, loss = 10.32 (290.1 examples/sec; 0.441 sec/batch) 2016-03-06 00:43:04.834940: step 110, loss = 10.27 (292.9 examples/sec; 0.437 sec/batch)
C-LIS CO., LTD. ػցֶशͷน σʔλͷน ܭࢉྔͷน ػցֶशͷน ܓೋʹػցֶश͕Θ͔Β͵ɻܓೋɺΞϓϦ։ൃऀͰ͋Δɻ+BWBΛॻ͖ɺ "OESPJE4UVEJPͱ༡ΜͰͯ͠དྷͨɻ͚ΕͲ؟ڸ່ͬʹରͯ͠ɺਓҰഒʹහײ
Ͱ͋ͬͨɻ
C-LIS CO., LTD. $*'"34BNQMF ଟΫϥεྨ IUUQTXXXDTUPSPOUPFEV_LSJ[DJGBSIUNM
C-LIS CO., LTD. ̎Ϋϥεྨ
C-LIS CO., LTD. ͬͨ͜ͱ ूΊͨσʔλΛ$*'"3ͷϑΝΠϧܗࣜʹ߹ΘͤͯQBDLͯ͠ αϯϓϧϓϩάϥϜʹ༩͑ͨ Label Red Green
Blue 1 0 Label Red Green Blue 1 Label Red Green Blue 2 :
C-LIS CO., LTD. DJGBSQZͷϞσϧ ̎ͷંΓΈɾϓʔϦϯά ̎ͷશ݁߹
C-LIS CO., LTD. ܇࿅σʔλ ςετσʔλ 5SBJOJOH CBUDI 4UFQ MOVING_AVERAGE_DECAY
= 0.9999 NUM_EPOCHS_PER_DECAY = 350.0 LEARNING_RATE_DECAY_FACTOR = 0.6 INITIAL_LEARNING_RATE = 0.2
C-LIS CO., LTD. 5SBJOJOH CBUDI 4UFQ
C-LIS CO., LTD. 5SBJOJOH CBUDI 4UFQ
C-LIS CO., LTD. &WBMVBUJPO ܇࿅σʔλ ςετσʔλ 1SFDJTJPO ʙ
ͦΕɺաֶश͡Όͳ͍ΜͰ͔͢ʁ
C-LIS CO., LTD. &WBMVBUJPO
C-LIS CO., LTD. ࠓޙͷ՝ ֶशʢςετʣσʔλͷॆ࣮ ʢ؟ڸͰͳ͘ɺΩϟϥΫλʔΛݟ͍ͯΔՄೳੑ͕͋Δʣ ఆਫ਼Λ্͛Δ ʢάϥϑͷਂԽʗύϥϝʔλʔௐʣ
C-LIS CO., LTD. ࠓޙͷ՝ ֶशʢςετʣσʔλͷॆ࣮ ʢ؟ڸͰͳ͘ɺΩϟϥΫλʔΛݟ͍ͯΔՄೳੑ͕͋Δʣ ఆਫ਼Λ্͛Δ ʢάϥϑͷਂԽʗύϥϝʔλʔௐʣ 8FCαʔϏεͱͯ͠ެ։
C-LIS CO., LTD. "OESPJEΞϓϦͷΈࠐΈ 5FOTPSqPX%FNP 1SFUSBJOFEʢ*ODFQUJPO.PEFMʣΛͬͨ ମೝࣝͷαϯϓϧΞϓϦ
C-LIS CO., LTD. %FNP
C-LIS CO., LTD. "OESPJEΞϓϦͷΈࠐΈ (SBEMFͰϏϧυͰ͖ͳ͍ʢ#B[FMʣ Ϟσϧσʔλ͕େ͖͍ ʢ*ODFQUJPOϞσϧͰ.#ʣ άϥϑΛؚΉϞσϧσʔλͷΤΫεϙʔτ͕खܰͰͳ͍ ·ͩૣ͍ɻ͚ΕͲࢹ͢Δඞཁ͕͋Δ
C-LIS CO., LTD. 5FOTPS'MPXΛͬͨػցֶश͜ͱ͡ΊCZ5PSV6&/0:"." IUUQXXXTMJEFTIBSFOFU5PSV6FOPZBNBUFOTPSqPXHEH 0QFO$7ʹΑΔΞχϝإݕग़ͳΒMCQDBTDBEF@BOJNFGBDF IUUQVMUSBJTUIBUFOBCMPHDPNFOUSZ σΟʔϓϥʔχϯάͰ͓ͦদ͞Μͷͭࢠݟ͚ΒΕΔͷ͔ʁɹʙ४උฤʙ IUUQCPIFNJBIBUFOBCMPHDPNFOUSZ $*'"3
$*'"3ͷσʔλΛಡΈࠐΉํ๏ IUUQRJJUBDPNTVQFSTBJBLVKJOJUFNTFECFG 5FOTPS'MPX IUUQTHJUIVCDPNUFOTPSqPXUFOTPSqPX ࢀߟ
(PPHMF5FOTPS'MPXͱ "OESPJE͕ܨ͕Δະདྷ
C-LIS CO., LTD. C-LIS CO., LTD.
ຊࢿྉɺ༗ݶձࣾγʔϦεͷஶ࡞Ͱ͢ɻܝࡌ͞Ε͍ͯΔΠϥετɺಛʹهࡌ͕ͳ͍߹ࠜઇΕ͍ͷஶ࡞Ͱ͢ɻ ຊࢿྉͷશ෦ɺ·ͨҰ෦ʹ͍ͭͯɺஶ࡞ऀ͔ΒจॻʹΑΔڐΛಘͣʹෳ͢Δ͜ͱې͡ΒΕ͍ͯ·͢ɻ 5IF"OESPJE4UVEJPJDPOJTSFQSPEVDFEPSNPEJpFEGSPNXPSLDSFBUFEBOETIBSFECZ(PPHMFBOEVTFE BDDPSEJOHUPUFSNTEFTDSJCFEJOUIF$SFBUJWF$PNNPOT"UUSJCVUJPO-JDFOTF ໊֤ɾϒϥϯυ໊ɺձ໊ࣾͳͲɺҰൠʹ֤ࣾͷඪ·ͨొඪͰ͢ɻຊࢿྉதͰɺɺɺäΛׂѪͯ͠ ͍·͢ɻ 5IF"OESPJESPCPUJTSFQSPEVDFEPSNPEJpFEGSPNXPSLDSFBUFEBOETIBSFECZ(PPHMFBOEVTFEBDDPSEJOH UPUFSNTEFTDSJCFEJOUIF$SFBUJWF$PNNPOT"UUSJCVUJPO-JDFOTF