Share
Embed
Transcript
42 slides
-
தՏ จ(Nakagawa Hirofumi) • ࢁݝग़ • ϝϧΧϦ20177݄ೖࣾ ex-MIRACLE LINUX, ex-Cerevo,
ex-mixi, ex-Drivemode(co-founder) • ॴଐSRE→AI/MLνʔϜ • σόΠευϥΠό։ൃ͔ΒϑϩϯτΤϯυ։ൃ·Ͱ ΔԿͰɻ࠷ۙͰMLͱRobotics͕ओઓͱͳΓ ͭͭ͋Δɻ Twitter: hnakagawa14 GitHub: hnakagawa 2 ࣗݾհ -
What is ࣸਅݕࡧ • ࣸਅݕࡧͱɺ͍ΘΏΔը૾ݕࡧػೳ • ΞϓϦ͔ΒࣸਅΛݩʹΛݕࡧ͢Δ • ໊ΛΒͳͯ͘ը૾͔ΒΛݕࡧ Ͱ͖Δ
4 ಈըϦϯΫ: https://youtu.be/kTni8EvOCgI -
جຊతͳࣸਅݕࡧͷΈ 5 Deep Neural Networks (DNN)Λ༻ͯ͠ը૾ ͔ΒಛϕΫτϧΛऔಘ औಘͨ͠ಛϕΫτϧΛ Approximate Nearest
Neighbor Index(ANN Index) ʹՃͯ͠ը૾indexΛߏங ݕࡧ࣌ʹಉ͘͡ը૾͔Β DNNΛհͯ͠ಛϕΫτϧΛऔ ಘ͠ɺANN Index͔Βݕࡧ 2 3 1 -
What is Kubernetes • KubernetesʢҎԼk8sʣͱΦʔϓϯιʔε ͷίϯςφɾΦʔέετϨʔγϣϯγες Ϝ • k8sʹCustom Resource
Definitionͱݺ ΕΔಠࣗͷϦιʔεΛఆٛͰ͖Δػೳ͕͋ Γɺ։ൃऀͦͷػೳΛհͯ͠k8sͷػೳΛ ֦ுͰ͖Δ • Amazon Elastic Container Service for Kubernetes (Amazon EKS) ͱk8sͷϚω ʔδυɾαʔϏεɺίϯτϩʔϧϓϨʔϯ ͷཧΛߦͬͯ͘ΕΔ 6 -
What is Custom Resource Definition • Custom Resource DefinitionʢҎԼCRDʣͱ ಠࣗʹϦιʔεΛఆٛͰ͖Δk8sͷػೳ
• CRDɾϦιʔεͱɺΧελϜɾίϯτϩʔ ϥͰߏ͞ΕΔ • ΧελϜɾίϯτϩʔϥ͕CRDɾϦιʔε ͷϥΠϑαΠΫϧ/ঢ়ଶʹԠͯ͡Ϋϥελͷ ঢ়ଶΛίϯτϩʔϧ͢Δ 7 -
ML Platform Lykeion ࣸਅݕࡧLykeionͱݺΕΔͷML Platform্ ʹߏங͞Ε͓ͯΓɺԼهͷػೳPlatformଆͷػೳ Λ༻͍ͯ͠Δ 8 • Training/Serving
CRD & ΧελϜίϯτϩʔϥ • ίϯςφϕʔεɾύΠϓϥΠϯ • Training/Serving ίϯςφΠϝʔδɾϏϧμʔ • ϞσϧɾϨϙδτϦ -
TrainingɾϦιʔεͷ࡞ • Training custom resourceΛCronJob͕࡞ • ΧελϜɾίϯτϩʔϥ͕CRDɾϦιʔε Ͱઃఆ͞ΕͨίϯςφϕʔεɾύΠϓϥΠ ϯΛ࣮ߦ •
࣮ߦ͢Δόον୯Ґͱͯ͠Hourly, Daily, Monthly͕ଘࡏ 12 -
ServingɾϦιʔεͷ࡞ • Image BuilderίϯςφɾΠϝʔδΛϏ ϧυͨ͋͠ͱɺServingΧελϜɾϦιʔ εΛ࡞ • ServingΧελϜɾίϯτϩʔϥCRDɾ ϦιʔεͷઃఆΛݩʹඞཁͳ DeploymentɺServiceΛ࡞
• ຊγεςϜͰߏங͞ΕͨANN IndexΛ ݸผͷIndexαʔϏεͱͯ͠σϓϩΠ 22 -
αʔϏεɾσΟεΧόϦ • Ϋϥελ্ʹଘࡏ͢ΔIndexαʔϏεΛ k8sΛհͯࣗ͠ಈతʹऔಘ͢Δ • ͳΔ͘େ͖ͳཻͷIndexΛ༻͢ΔΑ ͏ɺҟͳΔظؒɾཻͷIndexαʔϏε (Hourly, Daily, Monthly)
ΛࣗಈతʹΈ߹ ΘͤΔ • REST <-> IndexαʔϏεؒͷϓϩτίϧ gRPCΛ༻ 24 -
ࣸਅݕࡧͷόοΫΤϯυɾΠϯϑϥ 1. ίϯςφɾϕʔεͷ࠶ݱੑͷߴ͍γεςϜ 2. k8sͷCRD/ΧελϜɾίϯτϩʔϥαʔϏεɾσΟεΧόϦͷػೳΛ׆༻ 3. Batch Execution as Custom
ResourceɺML PlatformͰ࣮ݱ͞Ε͍ͯΔػೳΛ༻ ͠ɺϩόετͳγεςϜΛߏங 4. ΫϥυɾΠϯϑϥΛk8sͰநԽ͢ΔࣄʹΑͬͯɺ֤Ϋϥυɾϕϯμͷྑ͍ͱ͜औΓ Λ͍ͯ͠Δ 27 -
Realtime image search • ॴҦ Edge AI TechnologyΛ༻ͯ͠ɺݕࡧʹඞཁͳਪ ॲཧͷେ෦ΛEdgeଆͰߦ͍ͬͯΔ •
ϦΞϧλΠϜͳΠϯλϥΫγϣϯΛ࣮ݱ • UX্େ͖ͳϝϦοτ͕༗Δ 29 -
What must happen to make DNN work on edge •
༷ʑͳτϨʔυΦϑ͕ଘࡏ͢Δ • Accuracy • Latency • Energy consumption • Model size • తͷUXΛୡ͢ΔͨΊʹɺΞϧΰϦ ζϜɺΤϯδχΞϦϯά྆ํͰͦΕΒͷ όϥϯεΛߟྀ͢Δඞཁ͕͋Δ 31 Image credit: [1] Image credit: [1] Image credit: [2] ɾΦϖϨʔγϣϯʹΑͬͯίετ͕ҧ͏[1] ɾmobile deviceͰαϙʔτ͍ͯ͠ΔGPUΠϯλʔϑΣʔεͷγΣΞ[2]ɹ -
Landscape of execution environment Image credit: [2] Image credit: [2]
※ FacebookͷϨϙʔτ͔ΒͷҾ༻[2] ԿʹతͷUXΛ࣮ݱͰ͖ΔσόΠεͷΧόϨοδΛ͛Δ͔? 32 -
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] • ׆ੑԽؔʹh-swishΛ༻ • squeeze & excitationΛ ༻ͨ͠channelͷAttention • ܰྔͳfinal blockͷ࠾༻ etc... 33 -
Designing efficient networks: Automated ways - ௨ৗͷϞσϧͷτϨʔχϯά ΛτϨʔχϯάɾύϥϝʔλͱͨ͠Ϟσϧ Λೖྗɺ Λग़ྗͱͯ͠ɺ
Λ࠷খԽ͢Δ ୳ࡧ͢Δ - ΞʔΩςΫνϟɾαʔνͰͷ୳ࡧ&τϨʔχϯά ΞʔΩςΫνϟɾύϥϝʔλ Ճ Λ࠷খԽ͢Δ Λ୳ࡧ͢Δ ͱ 34 -
Two influential yet costly approaches 35 MnasNet[7] (RL-Based) FBNet[8] (Differentiable)
• ୳ࡧۭ͔ؒΒઍͷmodelΛsampling͢Δ • sample͞Εͨchild modelΛεΫϥον͔Β τϨʔχϯά͢Δ • ڊେͳ୳ࡧۭ͔ؒΒ୳ࡧͰ͖Δ͕ɺݱ࣮తͳ ΠςϨʔγϣϯΛߦ͏ҝʹɺڊେͳܭࢉػϦ ιʔε͕ඞཁʹͳΔ • DARTSϕʔεͷ୳ࡧख๏Λ࠾༻͍ͯ͠Δ • ୳ࡧۭؒͷ֤ΦϖϨʔγϣϯΛGPUϝϞϦʹ ͤΔඞཁ͕͋Δҝɺ݁ہGPUϝϞϦͷ༻ྔ͕ ͱͳΓɺsample͞Εͨproxy dataset͕ඞཁʹ ͳͬͨΓɺbatch sizeΛ্͛ΒΕͳ͔ͬͨΓ͢Δ Image credit: [7] Image credit: [8] -
Our approach 36 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๏ͰɺඦʙઍGPU͔͔࣌ؒ Δ୳ࡧ࣌ؒΛݮ͢Δ͜ͱ͕Ͱ͖Δ • MobileNet-V2ͱSingle-Path NAS(SPNAS)Ͱੜͨ͠Ϟσϧͱ ͷੑೳൺֱ MobileNet-V3Λϕʔεʹ୳ࡧۭؒΛઃఆ͠ɺSPNASͷϦʔζφϒϧͳNASख๏Ͱ୳ࡧ Λߦ͏ͷ͕ݱ࣮త? -
Backend System for Edge 37 Training Optimizer Distribute • Architecture
search • Weight pruning • Dense-Sparse-Dense training • Quantization • K-means cluster • Execution engine selection • Model version management • A/B testing • Distribute model ɾEdgeϓϩμΫτΫϥΠΞϯταΠυ͚ͩͰཱ͠ͳ͍ ɾRealtime ํϓϩτίϧ ɾDevice efficient ͳmodelΛੜ͢ΔΈ ɾmodelΛಈత͢ΔΈ etc.. -
Edge function architecture 38 38 Server side Client side •
αʔόଆͷDNN frameworkʹ TensorFlowΛ༻ • LykeionʹΑͬͯύΠϓϥΠϯΛ ߏங • ΫϥΠΞϯτଆͰTensorFlow Lite + MediaPipeΛ༻ • MediaPipeΛ༻͢Δ͜ͱͰલॲ ཧޙॲཧSIMDΛ༻ͯ͠ ޮԽͰ͖Δ (TF Lite) (Optional) -
Edge AI Technology 40 40 • ϦΞϧλΠϜͳΠϯλϥΫγϣϯΛ࣮ݱ͠UX্େ͖ͳϝϦοτ͕༗Δ • ͔͠͠ModelRuntimeɺ͞ΒʹBackendͱߟྀ͖͢ࣄฑ͕ଟ͘ͳΔͷࣄ࣮ •
తͷUXΛ࣮ݱ͢ΔͨΊʹɺ AccuracyLatencyͷόϥϯεΛऔΔඞཁ͕͋Δ • ඞͣ͠Accuracy͕࠷༏ઌͰͳ͍ • ࠓޙҰൠԽ͞ΕΔաఔͰɺҾ͖ଓ͖RuntimeModelingख๏ͷٸܹͳਐԽ͕༧͞ΕΔ -
References 41 [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).