メルカリ写真検索における Amazon EKS の活用事例と プロダクトにおけるEdgeAI technologyの展望

AWSLoftで行われた AWS Containers talk with Mercari で使用した資料です
https://awscontainertalkwithmercari.splashthat.com/

Transcript

1. © 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved. ϝϧΧϦࣸਅݕࡧʹ͓͚Δ Amazon EKS ͷ׆༻ࣄྫͱ
 ϓϩμΫτʹ͓͚ΔEdgeAI technologyͷల๬ !1 גࣜձࣾϝϧΧϦ தՏ ޺จ
2. © 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved. தՏ ޺จ • 2017೥7݄ೖࣾ • ॴଐ͸SRE→AI/MLνʔϜ • σόΠευϥΠό։ൃ͔ΒϑϩϯτΤϯυ ։ൃ·Ͱ΍ΔԿͰ΋԰ Twitter: hnakagawa14 GitHub: hnakagawa !2 ࣗݾ঺հ
3. © 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved. Introduction !3
4. © 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved. What is ࣸਅݕࡧ • ࣸਅݕࡧͱ͸ɺ͍ΘΏΔը૾ݕࡧػೳ • ΞϓϦ͔ΒࣸਅΛݩʹ঎඼Λݕࡧ͢Δ • ঎඼໊Λ஌Βͳͯ͘΋ը૾͔Β঎඼Λݕࡧ Ͱ͖Δ !4 ಈըϦϯΫ: https://youtu.be/kTni8EvOCgI
5. © 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved. جຊతͳࣸਅݕࡧͷ࢓૊Έ !5 Deep Neural Networks (DNN)Λ࢖༻ͯ͠঎඼ը૾ ͔Βಛ௃ϕΫτϧΛऔಘ औಘͨ͠ಛ௃ϕΫτϧΛ Approximate Nearest Neighbor Index(ANN Index) ʹ௥Ճͯ͠ը૾indexΛߏங ݕࡧ࣌ʹ͸ಉ͘͡঎඼ը૾͔Β DNNΛհͯ͠ಛ௃ϕΫτϧΛऔ ಘ͠ɺANN Index͔Βݕࡧ 2 3 1
6. © 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved. What is Kubernetes • KubernetesʢҎԼk8sʣͱ͸Φʔϓϯιʔε ͷίϯςφɾΦʔέετϨʔγϣϯγες Ϝ • k8sʹ͸Custom Resource Definitionͱݺ͹ ΕΔಠࣗͷϦιʔεΛఆٛͰ͖Δػೳ͕͋ Γɺ։ൃऀ͸ͦͷػೳΛհͯ͠k8sͷػೳΛ ֦ுͰ͖Δ • Amazon Elastic Container Service for Kubernetes (Amazon EKS) ͱ͸k8sͷϚω ʔδυɾαʔϏεɺίϯτϩʔϧϓϨʔϯ ͷ؅ཧΛߦͬͯ͘ΕΔ !6
7. © 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved. What is Custom Resource Definition • Custom Resource DefinitionʢҎԼCRDʣͱ ͸ಠࣗʹϦιʔεΛఆٛͰ͖Δk8sͷػೳ • CRDɾϦιʔεͱɺΧελϜɾίϯτϩʔ ϥͰߏ੒͞ΕΔ • ΧελϜɾίϯτϩʔϥ͕CRDɾϦιʔε ͷϥΠϑαΠΫϧ/ঢ়ଶʹԠͯ͡Ϋϥελͷ ঢ়ଶΛίϯτϩʔϧ͢Δ !7
8. © 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved. ML Platform Lykeion ࣸਅݕࡧ͸Lykeionͱݺ͹ΕΔ಺੡ͷML Platform্ ʹߏங͞Ε͓ͯΓɺԼهͷػೳ͸Platformଆͷػೳ Λ࢖༻͍ͯ͠Δ !8 • Training/Serving CRD & ΧελϜίϯτϩʔϥ • ίϯςφϕʔεɾύΠϓϥΠϯ • Training/Serving ίϯςφΠϝʔδɾϏϧμʔ • ϞσϧɾϨϙδτϦ
9. © 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved. Architecture !9
10. © 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved. Architecture֓ཁਤ !10 S3 EKS
11. © 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved. 1.TrainingɾϦιʔεͷ࡞੒ !11 S3 EKS
12. © 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved. TrainingɾϦιʔεͷ࡞੒ • Training custom resourceΛCronJob͕࡞੒ • ΧελϜɾίϯτϩʔϥ͕CRDɾϦιʔε Ͱઃఆ͞ΕͨίϯςφϕʔεɾύΠϓϥΠ ϯΛ࣮ߦ • ࣮ߦ͢Δόον୯Ґͱͯ͠͸Hourly, Daily, Monthly͕ଘࡏ !12
13. © 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved. ίϯςφϕʔεɾύΠϓϥΠϯ • ֤޻ఔΛݸผͷίϯςφɾΠϝʔδͰ࣮ߦ • ϥΠϒϥϦͷґଘؔ܎ͳͲ؀ڥφΠʔϒͳMLύΠϓϥΠϯͷ໰୊Λղܾ • ύΠϓϥΠϯDAG͸YAMLͰهड़ • ֤޻ఔͷೖग़ྗ͸Persistent VolumeʢҎԼPVʣΛհ͢ !13
14. © 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved. Batch Execution as Custom Resource • શͯͷόον࣮ߦ৘ใ͕CRDɾϦιʔεͱͯ͠ k8s্ʹ࢒Δ • ಉ͡ॲཧΛ࠶࣮ߦग़དྷΔͨΊɺόονͷ࠶࣮ߦ Λ൐͏ো֐෮چ࡞ۀ͕༰қ !14
15. © 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved. 2.ը૾ͷμ΢ϯϩʔυ !15 S3 EKS
16. © 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved. ը૾ͷμ΢ϯϩʔυ • Amazon S3্ʹଘࡏ͢ΔϝϧΧϦɾΠϝʔδετΞ͔Β঎඼ը૾Λμ΢ ϯϩʔυ • ύΠϓϥΠϯ্΋ͬͱ΋͕͔͔࣌ؒΔ޻ఔʢը૾਺͕๲େͳͨΊ) • ͦͷͨΊPVʹҰఆظؒΩϟογϡ͢ΔࣄʹΑͬͯ࠶ΠϯσοΫε͕ ඞཁͳ࣌ʹ͸ૉૣ͘ύΠϓϥΠϯΛճͤΔΑ͏ʹ͍ͯ͠Δ • PVͷ࣮ମ͸ Amazon EBS !16
17. © 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved. 3.ΞηοτͷΞοϓϩʔυ !17 S3 EKS
18. © 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved. ΞηοτͷΞοϓϩʔυ • ύΠϓϥΠϯͷ੒Ռ෺Ͱ͋Δಛ௃ϕΫτϧͱANN IndexΛϞσϧɾϨϙδτϦʹอଘ • શͯͷ੒Ռ෺͸όʔδϣϯ؅ཧ͞Εͨঢ়ଶͰอଘ͞ΕΔ • ϞσϧɾϨϙδτϦ͸GCS্ʹߏங !18
19. © 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved. 4.ServingΠϝʔδͷϏϧυ !19 S3 EKS
20. © 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved. ServingΠϝʔδͷϏϧυ 1. ϞσϧɾϨϙδτϦΛImage Builderͱݺ͹ΕΔdaemon͕؂ࢹ 2. ৽͍͠Serving͢΂͖Ϧιʔε͕௥Ճ͞ΕΔͱࣗಈͰServingίϯςφɾΠϝʔδΛϏϧυ • ίϯςφɾΠϝʔδ͸શͯͷANN Index౳ͷαʔϏϯάʹඞཁͳϦιʔεΛશؚͯΜͰ͍Δ 3. Ϗϧυ͞ΕͨίϯςφɾΠϝʔδΛίϯςφɾϨδετϦʹϓογϡ !20
21. © 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved. 5.ServingɾϦιʔεͷ࡞੒ !21 S3 EKS
22. © 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved. ServingɾϦιʔεͷ࡞੒ • Image Builder͸ίϯςφɾΠϝʔδΛϏ ϧυͨ͋͠ͱɺServingΧελϜɾϦιʔ εΛ࡞੒ • ServingΧελϜɾίϯτϩʔϥ͸CRDɾ ϦιʔεͷઃఆΛݩʹඞཁͳ DeploymentɺService౳Λ࡞੒ • ຊγεςϜͰ͸ߏங͞ΕͨANN IndexΛ ݸผͷIndexαʔϏεͱͯ͠σϓϩΠ !22
23. © 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved. 6.αʔϏεɾσΟεΧόϦ !23 S3 EKS
24. © 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved. αʔϏεɾσΟεΧόϦ • Ϋϥελ্ʹଘࡏ͢ΔIndexαʔϏεΛ k8sΛհͯࣗ͠ಈతʹऔಘ͢Δ • ͳΔ΂͘େ͖ͳཻ౓ͷIndexΛ࢖༻͢ΔΑ ͏ɺҟͳΔظؒɾཻ౓ͷIndexαʔϏε (Hourly, Daily, Monthly) Λࣗಈతʹ૊Έ߹ ΘͤΔ • REST <-> IndexαʔϏεؒͷϓϩτίϧ ͸gRPCΛ࢖༻ !24
25. © 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved. ֓ཁਤͷৼΓฦΓ !25 S3 EKS
26. © 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved. Conclusion !26
27. © 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved. ࣸਅݕࡧͷόοΫΤϯυɾΠϯϑϥ 1. ίϯςφɾϕʔεͷ࠶ݱੑͷߴ͍γεςϜ 2. k8sͷCRD/ΧελϜɾίϯτϩʔϥ΍αʔϏεɾσΟεΧόϦ౳ͷػೳΛ׆༻ 3. Batch Execution as Custom Resource౳ɺML PlatformͰ࣮ݱ͞Ε͍ͯΔػೳΛ࢖༻ ͠ɺϩόετͳγεςϜΛߏங 4. Ϋϥ΢υɾΠϯϑϥΛk8sͰந৅Խ͢ΔࣄʹΑͬͯɺ֤Ϋϥ΢υɾϕϯμͷྑ͍ͱ͜औΓ Λ͍ͯ͠Δ !27
28. © 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved. Amazon EKSʹ͍ͭͯ • k8sͱ͍͑͹GCPͱ͍͏Πϝʔδ͕͋Δ͕(։ൃݩ͔ͩΒ౰વ)ɺEKS΋͜ͳΕ͖ͯͯ҆ఆͨ͠ӡ ༻͕ग़དྷΔΑ͏ʹͳͬͨɻ͔͠͠Pros/Cons͸౰વଘࡏ͢Δɻ • Pros • ૉͷk8sʹۙ͘ɺͦ͏͍ͬͨҙຯͰ͸ॊೈͳߏ੒Λ૊Έ΍͍͢ • Cons • ଞࣾͷϚωʔδυk8sαʔϏεʹൺ΂ɺk8sʹ͍ͭͯඞཁͳࣄલ஌͕ࣝଟ͍ • ͔͠͠࢖͍উखͷྑ͍AWSͷS3΍RDS౳ͷଞαʔϏεΛɺk8sͱڞʹ࢖͍͍ͨϢʔβ͕᪳᪯͢ Δཧ༝͸΋͸΍ͳ͍ !28
29. © 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved. AWSͷྑ͍ͱ͜Ζ • AWSͷྑ͍ॴ͸Կͱݴͬͯ΋RDSɺAuroraʹ୅ද͞ΕΔɺϚωʔδυɾRDBMSαʔϏε • ࠷ॳRDSͰ࡞ͬͯɺύϑΥʔϚϯεʹࠔͬͨΒAuroraΛ࢖͏ • Persistent૚͕ॏཁͳɺಛʹRDBΛ࢖༻͢ΔγεςϜͰ͸ɺ࠷ॳͷબ୒ࢶʹͳΔ • ࠓճͷߏ੒ྫͱ͸গ͠ҧ͏͕ɺk8s(EKS)ͰϏδωεɾϩδοΫ͕ͷΔΠϯϑϥ෦෼Λ ந৅Խ͠Persistent૚ʹRDS౳ͷαʔϏεΛ࢖༻͢Δͷ͸͓͢͢Ίߏ੒ͷ̍ͭ !29
30. © 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved. Next Future? !30
31. © 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved. Realtime image search • ॴҦ Edge AI TechnologyΛ࢖༻ͯ͠ɺݕࡧʹඞཁͳਪ ࿦ॲཧͷେ෦෼ΛEdgeଆͰߦ͍ͬͯΔ • ϦΞϧλΠϜͳΠϯλϥΫγϣϯΛ࣮ݱ • UX্େ͖ͳϝϦοτ͕༗Δ !31
32. © 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved. Listing Dispacher • ద੾ͳग़඼ϝιουΛαδΣετͯ͘͠ΕΔ • ෳࡶͳग़඼ϑϩʔΛ؆ུԽ • ࠷ऴతʹ͸͔͚ͩ͟͢Ͱग़඼͕׬ྃ͢ΔॴΛ໨ࢦ͢!! !32
33. © 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved. What must happen to make DNN work on edge • ༷ʑͳτϨʔυΦϑ໰୊͕ଘࡏ͢Δ • Accuracy • Latency • Energy consumption • Model size • ໨తͷUXΛୡ੒͢ΔͨΊʹɺΞϧΰϦ ζϜɺΤϯδχΞϦϯά྆ํͰͦΕΒͷ όϥϯεΛߟྀ͢Δඞཁ͕͋Δ !33 Image credit: [1] Image credit: [1] Image credit: [2] ɾΦϖϨʔγϣϯʹΑͬͯίετ͕ҧ͏[1] ɾmobile deviceͰαϙʔτ͍ͯ͠ΔGPUΠϯλʔϑΣʔεͷγΣΞ[2]ɹ
34. © 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved. Landscape of execution environment Image credit: [2] Image credit: [2] ※ FacebookͷϨϙʔτ͔ΒͷҾ༻[2] ೗Կʹ໨తͷUXΛ࣮ݱͰ͖ΔσόΠεͷΧόϨοδΛ޿͛Δ͔? !34
35. © 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved. Designing efficient networks: Manual efforts - Mobile Nets V1, V2 and V3([4], [5], & [6]) • Depthwise separable conv Λ࢖༻͠ܭࢉྔΛ ௿ݮ • Inverted residual with linear bottleneck Λ࢖༻͠ ϝϞϦΞΫηεྔΛ ௿ݮ Image credit: [4] Image credit: [5] Image credit: [6] • ׆ੑԽؔ਺ʹh-swishΛ࢖༻ • squeeze & excitationΛ࢖ ༻ͨ͠channelͷAttention • ܰྔͳfinal blockͷ࠾༻ etc... !35
36. © 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved. Designing efficient networks: Automated ways - ௨ৗͷϞσϧͷτϨʔχϯά ΛτϨʔχϯάɾύϥϝʔλͱͨ͠Ϟσϧ Λೖྗɺ Λग़ྗͱͯ͠ɺ Λ࠷খԽ͢Δ ୳ࡧ͢Δ - ΞʔΩςΫνϟɾαʔνͰͷ୳ࡧ&τϨʔχϯά ΞʔΩςΫνϟɾύϥϝʔλ ௥Ճ Λ࠷খԽ͢Δ Λ୳ࡧ͢Δ ͱ !36
37. © 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved. Two influential yet costly approaches !37 MnasNet[7] (RL-Based) FBNet[8] (Differentiable) • ୳ࡧۭ͔ؒΒ਺ઍͷmodelΛsampling͢Δ • sample͞Εͨchild modelΛεΫϥον͔ ΒτϨʔχϯά͢Δ • ڊେͳ୳ࡧۭ͔ؒΒ୳ࡧͰ͖Δ͕ɺݱ࣮త ͳΠςϨʔγϣϯΛߦ͏ҝʹɺڊେͳܭࢉ ػϦιʔε͕ඞཁʹͳΔ • DARTSϕʔεͷ୳ࡧख๏Λ࠾༻͍ͯ͠Δ • ୳ࡧۭؒ಺ͷ֤ΦϖϨʔγϣϯΛGPUϝϞϦ ʹ৐ͤΔඞཁ͕͋Δҝɺ݁ہGPUϝϞϦͷ࢖ ༻ྔ͕໰୊ͱͳΓɺsample͞Εͨproxy dataset͕ඞཁʹͳͬͨΓɺbatch sizeΛ্͛ ΒΕͳ͔ͬͨΓ͢Δ Image credit: [7] Image credit: [8]
38. © 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved. Our approach !38 Single-Path NAS[9] Device SoC Generation (Snapdragon) Model ImageNet Top-1 Accuracy* Latency (ms)* A 845 SPNAS 74.48 77.90 A 845 MobileNetV2 71.80 76.36 B 808 SPNAS 73.07 113.92 B 808 MobileNetV2 71.80 162.82 C 670 SPNAS 73.15 92.14 C 670 MobileNetV2 71.80 111.85 D 801 SPNAS 71.93 84.65 D 801 MobileNetV2 71.80 120.82 Image credit: [9] * All results are for float32 • superkenelͱ͍͏୳ࡧۭؒͷઃఆ ख๏Ͱɺ਺ඦʙ਺ઍGPU͔͔࣌ؒ Δ୳ࡧ࣌ؒΛ࡟ݮ͢Δ͜ͱ͕Ͱ͖Δ • MobileNet-V2ͱSingle-Path NAS(SPNAS)Ͱੜ੒ͨ͠Ϟσϧͱ ͷੑೳൺֱ MobileNet-V3Λϕʔεʹ୳ࡧۭؒΛઃఆ͠ɺSPNAS౳ͷϦʔζφϒϧͳNASख๏Ͱ୳ࡧ Λߦ͏ͷ͕ݱ࣮త?
39. © 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved. Edge function architecture !39 39 Server side Client side • αʔόଆͷDNN frameworkʹ͸ TensorFlowΛ࢖༻ • LykeionʹΑͬͯύΠϓϥΠϯΛ ߏங • ΫϥΠΞϯτଆͰ͸TensorFlow Lite + MediaPipeΛ࢖༻ • MediaPipeΛ࢖༻͢Δ͜ͱͰલॲ ཧ΍ޙॲཧ΋SIMD౳Λ࢖༻ͯ͠ ޮ཰ԽͰ͖Δ (TF Lite) (Optional)
40. © 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved. Conclusion Again !40
41. © 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved. Edge AI Technology !41 41 • ϦΞϧλΠϜͳΠϯλϥΫγϣϯΛ࣮ݱ͠UX্େ͖ͳϝϦοτ͕༗Δ • ͔͠͠Model΍Runtimeɺ͞Βʹ͸Backendͱߟྀ͢΂͖ࣄฑ͕ଟ͘ͳΔͷ΋ࣄ࣮ • ໨తͷUXΛ࣮ݱ͢ΔͨΊʹɺ Accuracy΍Latency౳ͷόϥϯεΛऔΔඞཁ͕͋Δ • ඞͣ͠΋Accuracy͕࠷༏ઌͰ͸ͳ͍ • ࠓޙҰൠԽ͞ΕΔաఔͰɺҾ͖ଓ͖Runtime΍Modelingख๏ͷٸܹͳਐԽ͕༧૝͞ΕΔ
42. © 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved. References !42 [1] Lai, Liangzhen, Naveen Suda, and Vikas Chandra. "Not all ops are created equal!." arXiv preprint arXiv:1801.04326 (2018). [2] Wu, Carole-Jean, David Brooks, Kevin Chen, Douglas Chen, Sy Choudhury, Marat Dukhan, Kim Hazelwood et al. "Machine learning at facebook: Understanding inference at the edge." In 2019 IEEE International Symposium on High Performance Computer Architecture (HPCA), pp. 331-344. IEEE, 2019. [3] Howard, Andrew G., Menglong Zhu, Bo Chen, Dmitry Kalenichenko, Weijun Wang, Tobias Weyand, Marco Andreetto, and Hartwig Adam. "Mobilenets: Efficient convolutional neural networks for mobile vision applications." arXiv preprint arXiv:1704.04861 (2017). [4] Sandler, Mark, Andrew Howard, Menglong Zhu, Andrey Zhmoginov, and Liang-Chieh Chen. "Mobilenetv2: Inverted residuals and linear bottlenecks." In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 4510-4520. 2018. [5] Howard, Andrew, Mark Sandler, Grace Chu, Liang-Chieh Chen, Bo Chen, Mingxing Tan, Weijun Wang et al. "Searching for mobilenetv3." arXiv preprint arXiv:1905.02244 (2019). [6] Zoph, Barret, and Quoc V. Le. "Neural architecture search with reinforcement learning." arXiv preprint arXiv:1611.01578 (2016). [7] Tan, Mingxing, Bo Chen, Ruoming Pang, Vijay Vasudevan, Mark Sandler, Andrew Howard, and Quoc V. Le. "Mnasnet: Platform- aware neural architecture search for mobile." In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 2820-2828. 2019. [8] Wu, Bichen, Xiaoliang Dai, Peizhao Zhang, Yanghan Wang, Fei Sun, Yiming Wu, Yuandong Tian, Peter Vajda, Yangqing Jia, and Kurt Keutzer. "Fbnet: Hardware-aware efficient convnet design via differentiable neural architecture search." In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 10734-10742. 2019. [9] Stamoulis, Dimitrios, Ruizhou Ding, Di Wang, Dimitrios Lymberopoulos, Bodhi Priyantha, Jie Liu, and Diana Marculescu. "Single- path nas: Designing hardware-efficient convnets in less than 4 hours." arXiv preprint arXiv:1904.02877 (2019).
43. © 2019, Amazon Web Services, Inc. or its affiliates. All rights reserved. Thank you all for coming today !43