개발자가 알아두면 좋을 5가지 AWS 인공 지능 깨알 지식

개발자가 알아두면 좋을 5가지 AWS 인공 지능 깨알 지식

개발자가 알아두면 좋을 5가지 AWS 인공 지능 깨알 지식 - 윤석찬 (AWS 테크 에반젤리스트) from Amazon Web Services Korea

개발자가 알아두면 좋을 5가지 AWS 인공 지능 깨알 지식 - 윤석찬 (AWS 테크 에반젤리스트)

1. / // / © 2018, Amazon Web Services, Inc. or its Affiliates. All rights reserved.

2. 2,3 16 0 ( )

3. 1 .!

4. ( ) C A B ) https://aws.amazon.com/machine-learning/amis/ ) I PD L ) I PD L ) , ( )

5. . 45! F A 6A A 5 A: 6 : : F A A AF A A 6 6 6 c 23 T /-. d iN 1 A AF gh : K e f V I a X 7 L M b

6. AVX 512 72 vCPUs “Skylake” 144 GiB memory C5 12 Gbps to EBS 2X vCPUs 2X performance 3X throughput 2.4X memory C4 36 vCPUs “Haswell” 4 Gbps to EBS 60 GiB memory - ® 2 A I 1 5 / e V ! I l X 5

7. -

8. ! $aws ec2-run-instances ami-b232d0db --instance-count 1 --instance-type p2.8xlarge --region us-east-1 --user-data my_data_training.sh $aws ec2-stop-instances i-10a64379 A /

9. 2 .

10. 34 031 4 I 9 / F B G • 48313. 6 C 9 8 25 GP • ) 0 1. B C 6 C B 0 C • ,2 T 25 N • , T 25 I U V A( D 9B ,D 9B D 9B 0 - 4 4 () () - 2

11. 3 2 0 p3.2xlarge = $5 per hour p3.2xlarge x 20 = $100 per hour (0 ) 1

12. = $30 per hour

13. 45! 1 http://www.fast.ai/2018/08/10/fastai-diu-imagenet/ https://aws.amazon.com/blogs/machine-learning/new-speed-record-set-for-training-deep-learning-models-on-aws/ • ) ) 4 - $ 8 A • WB T P $ 0 • 09 36 1. 4 36 D (% NA

14. 1 4.75 8.5 12.25 16 1 4.75 8.5 12.25 16 Speedup(x) # GPUs Resnet 152 Inceptin V3 Alexnet Ideal P2.16xlarge (8 Nvidia Tesla K80 - 16 GPUs) Synchronous SGD (Stochastic Gradient Descent) 91% Efficiency 88% Efficiency 16x P2.16xlarge by AWS CloudFormation Mounted on Amazon EFS # GPUs P G A N M

15. https://nucleusresearch.com/research/single/guidebook-tensorflow-aws/ In analyzing the experiences of researchers supporting more than 388 unique projects, Nucleus found that 88 percent of cloud-based TensorFlow projects are running on Amazon Web Services. “

16. I P ( ) A !

17. .

18. , - H J N , - , - - A H https://aws.amazon.com/ko/sagemaker

19. .

20. K D ! https://medium.com/datadriveninvestor/auto-model-tuning-for-keras-on-amazon-sagemaker-plant-seedling-dataset-7b591334501e https://aws.amazon.com/blogs/machine-learning/amazon-sagemaker-automatic-model-tuning-produces-better-models-faster/ 248 9 = 6 1 C 7CG9 5 G 1 ) C 2C G 1 ) 0 $ ) 1 */ 348 ) 9 = 6 1 / C C 7CG9 5 G 1 * C 2C G 1 ( ) $ * 1 * ./ HN C ) (

21. S , 5 , I m z k kGe b M m k L a o A g sW M . 2 0 5 n . g k r A !

22. , .

23. © 김태훈, 책 읽어주는 딥러닝 https://carpedm20.github.io/tacotron/

24. , + + + © 김태훈, 책 읽어주는 딥러닝 https://carpedm20.github.io/tacotron/

25. -, - 2 5 ! • , A A • I A ) • ( ( P Matthew Takumi Vicki Aditi

26. 3 • ) 1 M 1 C ,, • A W C 1 ( ,. L S. 13 . !

27. ! 4 30 2 6 1 6, 1 M ) P5 5 ( , 2 - , 3 https://aws.amazon.com/ko/polly/pricing/

28. A K 1 0 % 4 I 4 2 6 D A 4 S 2 K K % 3 2 2% , / 9 5 K S 5 % %, 4 5 4 7 https://aws.amazon.com/ko/polly/customers/

29. A C • I • • I B • AP • • C • • - ! A

30. . Cloud Gem Portal Plugin - Text to Speech https://aws.amazon.com/ko/lumberyard/

31. .

32. / /. . 3 D A , / https://aws.amzon.com/ko/sumerian

33. / .

34. .

35. ! & ) OR A C ( C(/ , https://aws.amazon.com/ko/rekognition/

36. - "SourceImageFace": { "BoundingBox": { "Width": ..., "Height": ..., "Left": ..., "Top": ... }, "Confidence": 99.99964141845703 }, "FaceMatches": [ { "Similarity": 95, "Face": { "Landmarks": [ { "Type": "eyeLeft", "X": ... "Y": ... }, ... }, ] FaceMatches CompareFace

37. -

38. -

39. "Persons": [ { "Timestamp": number, "Person": { "Index": number, "BoundingBox": { "Width": number, "Top": number, "Height": number, "Left": number }, "Face": { "BoundingBox": { ... }, "Landmarks": { ... }, "Pose": { ... }, "Quality": { ... }, "Confidence": number } }, ... GetPersonTracking StartPersonTracking -

40. . Live Street Camera Amazon Kinesis Video Streams 1. Camera-captured video streams are processed by Kinesis Video Streams End User 3. End user is notified in case of face matches Amazon SNS AWS Lambda Amazon Kinesis Streams Amazon Rekognition Video Face collection 2. Rekognition Video analyses the video and searches faces on screen against a collection of millions of faces

41. T A T 8 E 2 T C R J C 0 S C N M 1 E E T P N https://aws.amazon.com/ko/rekognition/customers/

42. ! N dW g 3 , P WT l o n i P VL A 3 W a F , T d W Dl dS NC m VL A Amazon GuardDuty

43. -

44. -

45. . 2 Data Sources VPC flow logs DNS Logs CloudTrail Events Amazon CloudWatch rules Amazon GuardDuty AWS StepFunctions Lambda function End UserAmazon SNS 3. End user is notified in case of risk Lambda function EC2 Systems Manager EC2 2. EC2 System Manager fixes compromised EC2 Instances and credentials by documents 1. Guardduty continuously analyzes data sources and intelligently detect threats and sends CloudWatch Logs

46. !

47. 1 + 0 5 5

48. 1 + 0 5 5

49. KERAS AWS DEEP LEARNING AMI AMAZON SAGEMAKER REKOGNITION REKOGNITION VIDEO POLLY TRANSCRIBE TRANSLATE COMPREHEND LEX AWS DEEPLENS AMAZON MACHINE LEARNING SPARK & EMR AMAZON MECHANICAL TURK GPU ( P3 INSTANCES) CPU (C5) IOT (GREENGRASS) MOBILE G A E C

50. ! 4 4

51. • https://aws.amazon.com/ko/blogs/korea/deep-learning-on-aws-wanted-mathpresso-buzzbill/ • https://aws.amazon.com/ko/blogs/korea/hot-ai-startups-vuno-mathpresso-maru24-actionpower-elice/ C A C R Q &R O

52. , ?

53. , 2 .4 4 3 3 1 .

54. 43 2, 43 1 . .

55. ) 4A d ) 4A m I f W TRg TRg M a n o e ck i L à i lb TRg P o S (1 2 352 ( 2 ( 2 A ( 2 C (1 ( 2 2A ( 2 2 2A ( 2 2 4 3 ( 2 2 2 .

56. ! #

57. http://bit.ly/awskr-ml-credits L SW R . U - 12 U , ,W A , 12 http://bit.ly/awskr-feedback AWS Only

58. A . : S . : ./ / - . ::: - : -