Upgrade to Pro
— share decks privately, control downloads, hide ads and more …
Speaker Deck
Sign up for free
Menu
Search
Features
All features
Private URLs
Password Protection
Custom URLS
Scheduled publishing
Remove Branding
Restrict embedding
Deck Collections
Notes
Features
All features
Private URLs
Password Protection
Custom URLS
Scheduled publishing
Remove Branding
Restrict embedding
Deck Collections
Notes
Explore
Featured decks
Featured speakers
Programming
Technology
Storyboards
Explore
Featured decks
Featured speakers
Programming
Technology
Storyboards
Pricing
Search
Sign in
Sign up for free
20 minutes DeepLearning
Search
ryo nakamaru
December 08, 2016
Technology
530
0
Share
Embed
Copy iframe code
Copy JS code
Copy link
Start on current slide
20 minutes DeepLearning
re:Port 2016 での登壇資料です。
ryo nakamaru
December 08, 2016
More Decks by ryo nakamaru
See All by ryo nakamaru
AWSで楽をするサービスメッシュ入門/appmesh-trial
pottava
1
1.7k
reinforce-2019-recap-lt
pottava
2
4.2k
ScaleShift-jp-2019-summer
pottava
1
230
Firecracker とは何か/what is Firecracker
pottava
12
5.7k
ハイブリッド並列 on Kubernetes/hybrid-parallel-program-on-kubernetes
pottava
1
470
AWS Fargate + Code 兄弟で始める継続的デリバリー / Continuous Delivery with AWS Fargate and Code brothers
pottava
12
3.4k
Singularity と NVIDIA GPU Cloud で作る ハイブリッド機械学習環境の構築 / Building a hybrid environment for Machine Learning with Singularity and NGC
pottava
3
1.6k
明日から始めるちょい足し λ / get-started-with-aws-lambda
pottava
4
2.5k
NGC と Singularity によるハイブリッド機械学習環境 / A hybrid environment for Machine Learning with NGC and Singularity
pottava
0
550
Other Decks in Technology
See All in Technology
10分で知る最近のOmarchy
komagata
0
350
ADKで始める業務改善 - AIエージェント開発時の考えと設計
harappa80
2
150
AgentCore Runtime上にAgentic Coding基盤を構築・展開する際の設計ポイントと限界点 / Design considerations and limitations when building an agentic coding platform on AgentCore Runtime
har1101
4
200
アクセスキーこわい やめかたと漏らさない工夫
sassssan68
1
310
現場で役立つ技術負債の効果的な返済方法
masuda220
PRO
8
4.3k
なぜ「決定性」が 決定的に重要なのか? Durable Execution 基盤の数理的理解 #serverlessjp / ServerlessDays Tokyo 2026
ytaka23
0
240
銀行勘定系システムにおける開発プロセス刷新×AIによる環境モダナイゼーション / Development Process Transformation and AI-Driven Environment Modernization
muit
0
2.2k
Deployment の 先にある AI Agent 基盤 - kagent vNext、Agent Substrate、Hermes から読み解く Agent Runtime の現在地 / k8s-matsuri-2-ai-agent-platform-amsy810
masayaaoyama
3
610
え、こんなに早く改修できるの?──新人エンジニアとスクラムマスターの2人が語る、AI×アジャイル開発の現場
ysasago
0
140
10Xに技術的負債をもたらした「2つの境界の歪み」その構造と解消への営み
10xinc
0
2k
絵ではじめるKubernetesセキュリティ
aoi1
3
630
Goodbye ShellScript, Hello File-based App
shunsock
0
110
Featured
See All Featured
svc-hook: hooking system calls on ARM64 by binary rewriting
retrage
2
580
HDC tutorial
michielstock
2
870
Building the Perfect Custom Keyboard
takai
2
870
Leading Effective Engineering Teams in the AI Era
addyosmani
9
2.6k
The Director’s Chair: Orchestrating AI for Truly Effective Learning
tmiket
1
300
Ten Tips & Tricks for a 🌱 transition
stuffmc
1
230
New Earth Scene 8
popppiees
4
2.6k
YesSQL, Process and Tooling at Scale
rocio
174
15k
Dominate Local Search Results - an insider guide to GBP, reviews, and Local SEO
greggifford
PRO
0
330
The Psychology of Web Performance [Beyond Tellerrand 2023]
tammyeverts
49
3.6k
Refactoring Trust on Your Teams (GOTO; Chicago 2020)
rmw
35
3.8k
Building a Modern Day E-commerce SEO Strategy
aleyda
45
9.2k
Transcript
20 ͰΘ͔Δ DeepLearning AWS re:Invent ѲͷҰॿʹ re:Port 2016 @ 2016.12.07
தؙ ྑ @pottava SUPINF Inc.
ૣͰ͕͢ΫΠζͰ͢
re:Invent 2016 ϒϨΠΫΞτηογϣϯʁ
393 ʢAWS re:Invent2016ʲอଘ൛ʳશηογϣϯΛཏʂඇެࣜ·ͱΊ #reinvent ʛ Developers.IOʣ
re:Invent 2015, 2014 ϒϨΠΫΞτηογϣϯʁ
298, 214 ʢAWS re:Invent2015ʲอଘ൛ʳશηογϣϯΛཏʂඇެࣜ·ͱΊ #reinvent ʛ Developers.IOʣ ʢ2014ฐࣾͷϒϩάΑΓงғؾͰूܭʣ
Ͱ
re:Invent 2017 ϒϨΠΫΞτηογϣϯʁ
214 → 298 → 393 → ?
͜ΕΛػցʹ༧ଌͤ͞Δ ͦͷϓϩηε͕ػցֶश
͜ΕΛػցʹ༧ଌͤ͞Δ ͦͷϓϩηε͕ػցֶश
ຊ
20 ͰΘ͔Δ DeepLearning
20 ͰΘ͔Δ DeepLearning ͞ΘΓ͕
ΰʔϧ ɾձࣾʹ re:Invent ͷ͔͍͍ͬ͜ϨϙʔτΛग़ͤΔ ɾࠓճͷ Deep Learning ܥαʔϏεͳΒԶʹฉ͚ ɾࣾɺ࣌ AI
͢Α
ΰʔϧ ɾձࣾʹ re:Invent ͷ͔͍͍ͬ͜ϨϙʔτΛग़ͤΔ ɾࠓճͷ Deep Learning ܥαʔϏεͳΒԶʹฉ͚ ɾࣾɺ࣌ AI
͢Α ɹ΄Μͱʹݴ͏ͱΠλ͍
ࠓͷྲྀΕ 1. Deep Learning ͬͯͳΜͳͷʁ 2. ͦͷ্Ͱ AWS ͷൃදͬͯͲ͏ͩͬͨͷʁ
Deep Learning ͱ
ͷ·͑ʹ
Machine Learning ͱ
ػցֶशͷେલఏ ػցֶशʹ 2 ͭͷϑΣʔζ͕͋Γ·͢ɻ ͦΕ ֶशɹͱɹ ਪ
ֶशͱ ΑΓ͏·͘ʹରॲͰ͖ΔΑ͏ʹͳΔաఔͷ͜ͱɻ
͡Ό͋ਪʁ ֶΜͩ͜ͱΛͬͯΛղ͘͜ͱɻ
͜Εֶश
͜Εਪ
͜Εֶश
͜Εਪ དྷ re:Invent ߦ͖͍ͨΜ͚ͩͲ ձࣾʹԿΛใࠂ͢Δͷ͕ਖ਼ղʁ
େࣄͳ͜ͱͳͷͰ͏Ұ ػցֶशͷ 2 ͭͷϑΣʔζɿ ֶशɹͱɹ ਪ
ͳΔ΄Ͳ
ػցֶशʢMachine Learningʣͱ ػցʹֶशͤ͞ɺਪͤ͞Δ͜ͱɻ ղ͖͍ͨ࣍ୈͰɺֶशɾਪͷํ๏Λม͑Δɻ ɾճؼ ɾΫϥεྨ ɾΫϥελϦϯά ɾ..
Deep Learning ͱ
DeepLearningͱ
DeepLearningͱ ͳΔ΄Ͳɺ͍͍
Կͷਂ͍ϨΠϠʔʢχϡʔϥϧωοτϫʔΫʣΛ ͬͯɺֶशɾਪ͢Δػցֶशͷ͜ͱɻ ϨΠϠʔ͕ਂ͍͔ΒσΟʔϓɻ ຊޠͳΒɺਂֶशɻਂ͍ɻ DeepLearningͱ
ػցֶशʢMLʣͱਂֶशʢDLʣ ML DL ɾ౷ܭతੳʢճؼɺΫϥεྨɾɾʣ ɾਐԽతܭࢉ ɾχϡʔϥϧωοτϫʔΫ ɾ… ɾΈࠐΈχϡʔϥϧωοτϫʔΫʢCNNʣ ɾ࠶ؼܕχϡʔϥϧωοτϫʔΫʢRNNʣ ɾਂ
Q ωοτϫʔΫʢDQNʣ ɾ…
ɾAlphaGo ɾ?? ɾ?? DeepLearning Կ͕͍͢͝ͷʁ
AlphaGo ਓؒͷੈքҰʹίϯϐϡʔλ͕উͬͯ͠·ͬͨɾɾʂ ͍ͭ࠷ۙ·ͰɺAI ع࢜ʹউͯͳ͍ɺͦͯ͠ϒϨΠΫεϧʔʹ 10 ͨͶͳΒͳ͍ͱݴΘΕ͍ͯͨͷʹɺͩɻ
ʢจ: http://www.nature.com/nature/journal/v529/n7587/abs/nature16961.html ʣ • 4ͭͷσΟʔϓχϡʔϥϧωοτϫʔΫΛར༻ • ͦΕͧΕ ॳڃऀpπ / தڃऀpσ
/ ্ڃऀpρ / அvθ ͱ͍ͬͨҐஔ͚ • pπ ਓؒͷ800ສͷ൫໘σʔλΛݩʹֶशɻਫ਼͍͕ߴʹղΛಘΔɻ • pσ 13ͷCNNɻ3,000ສ൫໘Λ50GPUͰ3.4ԯεςοϓɺ3िֶؒशɻϓϩͷࢦ ͠खΛ57.0%ͷਫ਼Ͱ༧Ͱ͖Δɻ • pρ 50GPUͰ1͔͚128ສճࣗݾରઓɻطଘιϑτʹ85%ͷѹతউɻ • vθ pσ ͰϥϯμϜʹ3,000ສ൫໘Λੜ͠ɺpρ Ͱ1ԯ6,000ສճϩʔϧΞτͨ͠ উΛڭࢣσʔλͱ͠ɺ50GPUͰҰिؒ 5,000ສճ֬ޯ߱Լ๏Λ࣮ࢪɻ • ࣮ରઓͰ1,202CPU+176GPU͕ΘΕɺpσͰ࣍ͷखબɺvθͰ൫໘ධՁɻ • উ͍͍͕ཧ٧ΊͷpρΑΓɺਓؒͷบΛֶΜͩpσΛ͔ͬͨɻ • উҼ pρ ͷపఈతͳڧԽֶशʹՃ͑ɺϞϯςΧϧϩ୳ࡧͱCNNͷ߹͔ͤɻ AlphaGo
ɾAlphaGo ɾAmazon Go ɾ?? DeepLearning Կ͕͍͢͝ͷʁ
Amazon Go 1. ೖళ࣌ɺήʔτʹεϚϗΞϓϦΛ͔͟͢ 2. ΄͍͠ͷΛόοάʹೖΕΔ 3. ͓ళΛग़Δ Coming early
2017 !! 2131 7th Ave Seattle, Washington
Amazon Go
Amazon Go ళ಄Ͱɺਪ͑͞ಈ͚͍͍ɻ ͍ɺͦΜͳ͜ͱΑΓ ͏ͪͷళฮʹ͜Εಋೖͯ͠Έ͍ͨΜͰ͚͢Ͳɾɾʂ
OK
ਂֶशϕʔεͷը૾ೝࣝ APIɻ ɾҰൠମ / ܠݕग़ ɾදੳ ɾإͷྨࣅఆ
Amazon Rekognition ଞࣾͰΜͳ Computer vision API ͷҰछɻ
Amazon Rekognition ฐࣾ༐ऀͷ ྨࣅఆɻ
ଓ͖·ͯ͠
ɾAlphaGo ɾAmazon Go ɾAmazon Echo DeepLearning Կ͕͍͢͝ͷʁ
Amazon Echo Alexa ʹԻͰ͓ئ͍ɾ࣭͢ΔͨΊͷσόΠεɻ ʢAlexa Amazon ͕։ൃͨ͠ AI ʣ
ʮΞϨΫαɺిؾΛ͚ͭͯɺԻָ͔͚ͯʯ ʮΞϨΫαɺUber ΛݺΜͰɻࠓͷఱؾʁ ʯ
Amazon Echo
Amazon Echo 1. ԻΛฉ͖औΓ 2. ԿΒ͔ͷॲཧΛͯ͠ 3. ԻΛฦ͢
Amazon Echo ɾӡసதͷϋϯζϑϦʔΞϓϦ࡞Γ͍ͨ ɾԻग़ྗɺTwillio ʹ͑ͦ͏ͳ ɾ..
OK
Amazon Polly ςΩετΛԻʹม͢Δ APIɻ ɾAlexa ͷԠػೳʮग़ൃ͍ͭͰ͔͢ʁʯ ɾຊޠʹରԠͯ͠Δʂ ɾXML ͰͤԻͷΧελϚΠζ͔ΜͨΜ
Amazon Lex ରΠϯλʔϑΣΠε + ࣗવݴޠղੳ APIɻ ɾϢʔβͱର͠ͳ͕Β࣍ͷॲཧΛஅ ɾ೦ͳ͕ΒɺݱࡏӳޠͷΈͷରԠ
Amazon Echo 1. ԻΛฉ͖औΓ 2. ԿΒ͔ͷॲཧΛͯ͠ 3. ԻΛฦ͢ Amazon Lex
Amazon Polly
ͪͳΈʹଞࣾͷঢ়گ
ɾVision API ɹը૾ೝࣝ APIɻମ / จࣈ / إ / ϩΰ
/ ϥϯυϚʔΫͷݕɻ ɾSpeech API ɹಉࣾͷར༻͢ΔԻʹΑΔςΩετೖྗΛ REST API Խɻ ɾNatural Language API ɹจষΛॻ͍ͨਓͷҙݟײɺݻ༗໊ࢺ࣌ؒදݱΛநग़ɻ ɾTranslate API ɹઍͷݴޠؒͰϦΞϧλΠϜʹςΩετΛ༁ɻ https://cloud.google.com/#products_machine-learning
ɾVision ɹը૾ೝࣝɾՃɺײੳɺإݕग़ɾಉҰఆɾྨࣅݕࡧɺಈըੳͳͲɻ ɾSpeech ɹจষͱԻͷ૬ޓมɺԻʹΑΔϢʔβࣝผɾೝূɻ ɾLanguage ɹݴޠɾτϐοΫɾΩʔϑϨʔζநग़ɺωΨϙδఆɺεϖϧνΣοΫɺ༧ଌɻ ɾKnowledge ɹࣝσʔλϕʔεΛ༻͍ͨੳɺ༧ଌɻ ɾSearch ɹWebɾಈըɾը૾ɾχϡʔεɾؔ࿈ϫʔυͷݕࡧɾऩूͳͲɻ
https://azure.microsoft.com/ja-jp/services/cognitive-services/
ɾLanguage ɹࣗવݴޠॲཧʹΑΔςΩετղੳɻಛఆใɾΩʔϫʔυɾίϯηϓτநग़ɺ ɹΧςΰϦྨɺײੳɺωΨϙδఆɺެจղੳɾҙຯ͚Ԡɻ ɾSpeech ɹจষͱԻͷ૬ޓมɻ ɾVision ɹը૾ೝࣝɻ ɾData Insights ɹχϡʔεϒϩάهࣄͷݕࡧɺτϨϯυੳɻ
ɹաͳબࢶͷআ֎ɺෳͷໃ६͢ΔΰʔϧΛؚΉΑ͏ͳෳࡶͳܾఆͷੳɻ https://www.ibm.com/watson/developercloud/services-catalog.html
ͳΔ΄ͲͰ͢Ͷɻ
ͯ͞ AWS ͷ 3 ͭͷ৽αʔϏε
Rekognition / Lex / Polly ͷڞ௨ શ෦ɺਪ API
ਪͷαʔϏεͱ͍͏͜ͱ ֶश AWS ͤɻ͜Μͳ͜ͱʹ͍͍ͯͳ͍ ɾࣗࣾͷΛը૾ݕग़͢Δ ɾࣗࣾͷΩϟϥΫλʔޠʁΛղऍͤ͞Δ ɾࣗࣾήʔϜͷຐ๏Λࣗવݴޠղੳͤ͞Δ
͑ʔ
ࣗࣾͷѻ͍͍ͨͳʔ
ํͳ͍ɺֶश͠Α͏
Α͠ɺֶशͩ 0. ͦΕɺDeepLearning Ͱղ͘ͷ͕࠷దͳͷʁ 1. σʔλ͕ඞཁɻʢ2016 ݱࡏɺҰൠతʹʣͨ͘͞Μ 2. ֶशΞϧΰϦζϜͷબ 3.
ܭࢉ͠·͘Δίϯϐϡʔλࢿݯ͕ඞཁ
Α͠ɺֶशͩ 0. ͦΕɺDL Ͱղ͘ͷ͕࠷దͳͷʁ౷ܭతੳʁ 1. σʔλ͕ඞཁɻͨ͘͞Μ 2. ֶशΞϧΰϦζϜͷબ 3. ܭࢉ͠·͘Δίϯϐϡʔλࢿݯ͕ඞཁ
σʔλΛूΊΑ͏ ʢ͜ͷલʹ͖͢ɺେࣄͳॾݕ౼ AWS αʔϏεؔͳ͍ͷͰஔ͍ͱ͍ͯʣ σʔλΛूΊͳ͚Ε࢝·Βͳ͍ʂ ɾKinesis Streams ɾCloudWatch Logs ɾS3
/ RDS / DynamoDB
ڭࢣσʔλΛ࡞Ζ͏ ·ͩಈը / εϥΠυͳ͔͚ͬͨͲɾɾ ϝΧχΧϧλʔΫΛ͏͕͋Γ·ͨ͠ɻ
ֶशϑϨʔϜϫʔΫΛબ΅͏ ͲΕ͕͍͍ΜͩΖ͏ɾɾʁ MXNet / TensorFlow / Caffe / Chainerɻ ɾͲͷχϡʔϥϧωοτ͏ͷʁCNNʁRNNʁ
ɾGPU ͏ͷʁCPU ͚ͩʁෳϊʔυల։ඞཁʁ ɾࠃ࢈Ԡԉͪ͠Ό͏ʁ ɾ..
AWS MXNet Ұͩͦ͏Ͱ͢ ɾ͑ɺAmazon DSSTNE ɾɾ ɾଞͷͷݕ౼͍ͨ͠ํͪ͜Β ɹ re:Invent
2016 Bringing Deep Learning to the Cloud with Amazon EC2 (CMP314)
Deep Dive ͍ͨ͠ํ
ࠓͷΛΑΓਂ͘ɺAI ؔ࿈ใ͚ͩͷ 120 ʂʂ http://jawsug-ai.connpass.com/
࠷ޙʹɺֶशΠϯϑϥબఆ
DL ͚ AWS ϚωʔδυαʔϏε దࡐదॴͰϐλΰϥεΠονɺࠓޙ͋Γ͑Δ ɾGlue ɾEMR ɾECS ɾBatch ɾ..
new! new!
͑ɺAmazon ML ʁ ML DL ͜͜ͷྖҬͰΛղ͘ͳΒͪΖΜ͑·͢ʂ Amazon ML
Amazon ECS ɾre:Invent Ͱͷ MXNet ϫʔΫγϣοϓָ͔ͬͨ͠ ɹhttps://github.com/awslabs/ecs-deep-learning-workshop/ ɾECS Ϋϥελ্Ͱ MXNet
ͷֶशɾਪ ɾDocker ίϯςφલఏʹͳΔ͕ ɾDocker ༷ʑϝϦοτ͕͋Γͭͭ ɾίϯςφΫϥελͱͯ͠ಈ͔͢ͳΒ࠷ߴ
AWS Batch ɾECS ্ʹ HPC ۀքͷҙຯʹ͍ۙΫϥελΛߏஙɻ ɾδϣϒεέδϡʔϥ ≠ ίϯςφք۾ͷεέδϡʔϥ ɾECS
্ͳͷͰɺίϯςφϕʔε ɾGlue EFS ͱͷΈ߹Θ͕ͤॏཁͦ͏
ECS ͱ Batch ɾֶशϑΣʔζɺେྔͷσʔλͱେྔͷܭࢉ - EFS / EBS / Batch
ͳͲʹΑΔδϣϒཧ ɾਪϑΣʔζɺେྔͷϦΫΤετΛࡹ͘ - Մ༻ੑߴ͘εέʔϧՄೳͳৗ࣌ىಈ API αʔό - ECS ্ʹαʔϏεσϓϩΠ
Deep Dive ͢ΔͳΒ
ECS Λத৺ʹɺίϯςφ·ΘΓͷ࠷৽ใΛ͓ಧ͚ʂ http://jawsug-container.connpass.com/
http://prtimes.jp/main/html/rd/p/000000007.000007768.html Comfy for Docker ϓϩδΣΫτͷ Docker ಋೖɾ։ൃࢧԉɾӡ༻ࢹߦΛ͍ͨ͠·͢ɻ https://www.supinf.co.jp/service/dockersupport/
ͪͳΈʹଞࣾͷঢ়گ
GPU ΛͬͨϑϧϚωʔδυͳ TensorFlowɻ Google ͷάϩʔόϧ༧ଌϓϥοτϑΥʔϜͰઍϢʔβɺTB ڃσʔλͰɻ ͪΖΜɺCloud Storage BigQuery
ͱͷ౷߹؆୯ɻ TensorFlow ͕ۤखͳॲཧɾɾ
GPUInfinibandΛؚΉϚϧνΫϥυHPCϋʔυΣΞ͕͑Δɻ ΦϯϓϨϛεͱϋΠϒϦουԽͰ͖ɺ౷߹͞Εͨ160Λ͑ΔιϑτΣΞ ύοέʔδ͕؆୯ʹ͑ɺ ैྔ՝ۚܕͰͪ࣌ؒͳ͘ HPCϦιʔεΛ͙͢ʹɻ
AI ָ͍͠Ͱ͢Ͷʂʂ
ΈΜͳͰ AWS ʹϑΟʔυόοΫΛʂ
͓ΘΓ
גࣜձࣾεϐϯϑ ΞΠσΟΞΛ͔ͨͪʹʂ +
͝૬ஊ͓ؾܰʹͪ͜Β·Ͱ.. 93 <Thank you !! https://www.supinf.co.jp/service/dockersupport/