![Page 2: Introduction to Amazon Sagemaker - Seoul AIseoulai.com › presentations › Introduction_to_Amazon_Sagemaker.pdf · Benefits of using machine learning in cloud 1. Size of dataset](https://reader030.vdocuments.mx/reader030/viewer/2022041022/5ed3bde1e4e8c10a396fae01/html5/thumbnails/2.jpg)
Benefits of using machine learning in cloud1. Size of dataset
2. Clustered training
3. Hyperparameter tuning
4. Training / Serving Skew
5. De-coupling model from client
6. Autoscaling prediction
Source: https://www.udemy.com/from-0-to-1-tensorflow-for-deep-learning/learn/v4/overview
![Page 3: Introduction to Amazon Sagemaker - Seoul AIseoulai.com › presentations › Introduction_to_Amazon_Sagemaker.pdf · Benefits of using machine learning in cloud 1. Size of dataset](https://reader030.vdocuments.mx/reader030/viewer/2022041022/5ed3bde1e4e8c10a396fae01/html5/thumbnails/3.jpg)
Source: https://www.slideshare.net/AmazonWebServices/new-launch-integrating-amazon-sagemaker-into-your-enterprise-mcl345-reinvent-2017
Amazon SageMakerA fully managed service that enables data scientists and developers to quickly and easily
build machine-learning based models into production smart applications.
Notebook Instances 1P Algorithms ML Training Service ML Hosting Service
![Page 4: Introduction to Amazon Sagemaker - Seoul AIseoulai.com › presentations › Introduction_to_Amazon_Sagemaker.pdf · Benefits of using machine learning in cloud 1. Size of dataset](https://reader030.vdocuments.mx/reader030/viewer/2022041022/5ed3bde1e4e8c10a396fae01/html5/thumbnails/4.jpg)
Noteboook Instance Few clicks
![Page 5: Introduction to Amazon Sagemaker - Seoul AIseoulai.com › presentations › Introduction_to_Amazon_Sagemaker.pdf · Benefits of using machine learning in cloud 1. Size of dataset](https://reader030.vdocuments.mx/reader030/viewer/2022041022/5ed3bde1e4e8c10a396fae01/html5/thumbnails/5.jpg)
Noteboook Instance Get a jupyter notebook
![Page 6: Introduction to Amazon Sagemaker - Seoul AIseoulai.com › presentations › Introduction_to_Amazon_Sagemaker.pdf · Benefits of using machine learning in cloud 1. Size of dataset](https://reader030.vdocuments.mx/reader030/viewer/2022041022/5ed3bde1e4e8c10a396fae01/html5/thumbnails/6.jpg)
Noteboook Instance
Pros● Easy access to data in cloud● Web interface● No installation of packages
Cons● Cost
○ Free tier for 2 month● Region limitation
○ Seoul is not supported● Package version constraints
![Page 7: Introduction to Amazon Sagemaker - Seoul AIseoulai.com › presentations › Introduction_to_Amazon_Sagemaker.pdf · Benefits of using machine learning in cloud 1. Size of dataset](https://reader030.vdocuments.mx/reader030/viewer/2022041022/5ed3bde1e4e8c10a396fae01/html5/thumbnails/7.jpg)
1P Algorithms
Source: https://www.slideshare.net/AmazonWebServices/new-launch-introducing-amazon-sagemaker-mcl365-reinvent-2017
![Page 8: Introduction to Amazon Sagemaker - Seoul AIseoulai.com › presentations › Introduction_to_Amazon_Sagemaker.pdf · Benefits of using machine learning in cloud 1. Size of dataset](https://reader030.vdocuments.mx/reader030/viewer/2022041022/5ed3bde1e4e8c10a396fae01/html5/thumbnails/8.jpg)
ML Training Service
Source: https://www.slideshare.net/AmazonWebServices/new-launch-infinitely-scalable-machine-learning-algorithms-with-amazon-ai-mcl341-reinvent-2017
Cost vs Time
![Page 9: Introduction to Amazon Sagemaker - Seoul AIseoulai.com › presentations › Introduction_to_Amazon_Sagemaker.pdf · Benefits of using machine learning in cloud 1. Size of dataset](https://reader030.vdocuments.mx/reader030/viewer/2022041022/5ed3bde1e4e8c10a396fae01/html5/thumbnails/9.jpg)
Distributed training1. Scale to hundreds of GPU2. Train outrageously large models3. Customize models with few lines of codes
But…
Need to manage instance orchestration (Kubernetes or Mesos)
Need to manage fault tolerance
An example of managing kubernetes cluster https://blog.openai.com/infrastructure-for-deep-learning/
Source: https://youtu.be/la_M6bCV91M?list=PLOU2XLYxmsIKGc_NBoIhTn2Qhraji53cv
![Page 10: Introduction to Amazon Sagemaker - Seoul AIseoulai.com › presentations › Introduction_to_Amazon_Sagemaker.pdf · Benefits of using machine learning in cloud 1. Size of dataset](https://reader030.vdocuments.mx/reader030/viewer/2022041022/5ed3bde1e4e8c10a396fae01/html5/thumbnails/10.jpg)
Managed vs manual
Source: https://youtu.be/la_M6bCV91M?list=PLOU2XLYxmsIKGc_NBoIhTn2Qhraji53cv&t=1358
![Page 11: Introduction to Amazon Sagemaker - Seoul AIseoulai.com › presentations › Introduction_to_Amazon_Sagemaker.pdf · Benefits of using machine learning in cloud 1. Size of dataset](https://reader030.vdocuments.mx/reader030/viewer/2022041022/5ed3bde1e4e8c10a396fae01/html5/thumbnails/11.jpg)
Source: https://www.slideshare.net/AmazonWebServices/new-launch-introducing-amazon-sagemaker-mcl365-reinvent-2017
![Page 12: Introduction to Amazon Sagemaker - Seoul AIseoulai.com › presentations › Introduction_to_Amazon_Sagemaker.pdf · Benefits of using machine learning in cloud 1. Size of dataset](https://reader030.vdocuments.mx/reader030/viewer/2022041022/5ed3bde1e4e8c10a396fae01/html5/thumbnails/12.jpg)
ML Training Service
Source: https://youtu.be/t64ortpgS-E?list=PLOU2XLYxmsIKGc_NBoIhTn2Qhraji53cv&t=361
To use cloud services tf.estimator is
required
![Page 13: Introduction to Amazon Sagemaker - Seoul AIseoulai.com › presentations › Introduction_to_Amazon_Sagemaker.pdf · Benefits of using machine learning in cloud 1. Size of dataset](https://reader030.vdocuments.mx/reader030/viewer/2022041022/5ed3bde1e4e8c10a396fae01/html5/thumbnails/13.jpg)
ML Training Service Build models by using estimators
![Page 14: Introduction to Amazon Sagemaker - Seoul AIseoulai.com › presentations › Introduction_to_Amazon_Sagemaker.pdf · Benefits of using machine learning in cloud 1. Size of dataset](https://reader030.vdocuments.mx/reader030/viewer/2022041022/5ed3bde1e4e8c10a396fae01/html5/thumbnails/14.jpg)
ML Training Service Need to wrap with sagemaker api
Version constraints ( Python 2.7, Tensorflow 1.4.0 )Source: https://github.com/aws/sagemaker-python-sdk
![Page 15: Introduction to Amazon Sagemaker - Seoul AIseoulai.com › presentations › Introduction_to_Amazon_Sagemaker.pdf · Benefits of using machine learning in cloud 1. Size of dataset](https://reader030.vdocuments.mx/reader030/viewer/2022041022/5ed3bde1e4e8c10a396fae01/html5/thumbnails/15.jpg)
ML Training Service Training PyTorch is possible
Source: https://www.slideshare.net/AmazonWebServices/new-launch-introducing-amazon-sagemaker-mcl365-reinvent-2017
![Page 16: Introduction to Amazon Sagemaker - Seoul AIseoulai.com › presentations › Introduction_to_Amazon_Sagemaker.pdf · Benefits of using machine learning in cloud 1. Size of dataset](https://reader030.vdocuments.mx/reader030/viewer/2022041022/5ed3bde1e4e8c10a396fae01/html5/thumbnails/16.jpg)
ML Training Service
Summary● Easy to use distributed training
● Supporting various frameworks
○ Tensorflow
○ MxNet
○ PySpark
○ PyTorch* …
[*]: BYO Docker Containers with SageMaker Estimators
![Page 17: Introduction to Amazon Sagemaker - Seoul AIseoulai.com › presentations › Introduction_to_Amazon_Sagemaker.pdf · Benefits of using machine learning in cloud 1. Size of dataset](https://reader030.vdocuments.mx/reader030/viewer/2022041022/5ed3bde1e4e8c10a396fae01/html5/thumbnails/17.jpg)
Source: https://www.slideshare.net/AmazonWebServices/new-launch-introducing-amazon-sagemaker-mcl365-reinvent-2017
![Page 18: Introduction to Amazon Sagemaker - Seoul AIseoulai.com › presentations › Introduction_to_Amazon_Sagemaker.pdf · Benefits of using machine learning in cloud 1. Size of dataset](https://reader030.vdocuments.mx/reader030/viewer/2022041022/5ed3bde1e4e8c10a396fae01/html5/thumbnails/18.jpg)
ML Hosting Service
Summary● Easy to deploy
● Auto-Scaling Inference APIs
● A/B Testing
● Low Latency & High Throughput
● Bring Your Own Model
● Python SDK
![Page 19: Introduction to Amazon Sagemaker - Seoul AIseoulai.com › presentations › Introduction_to_Amazon_Sagemaker.pdf · Benefits of using machine learning in cloud 1. Size of dataset](https://reader030.vdocuments.mx/reader030/viewer/2022041022/5ed3bde1e4e8c10a396fae01/html5/thumbnails/19.jpg)
Source: https://www.altexsoft.com/blog/datascience/comparing-machine-learning-as-a-service-amazon-microsoft-azure-google-cloud-ai/
![Page 20: Introduction to Amazon Sagemaker - Seoul AIseoulai.com › presentations › Introduction_to_Amazon_Sagemaker.pdf · Benefits of using machine learning in cloud 1. Size of dataset](https://reader030.vdocuments.mx/reader030/viewer/2022041022/5ed3bde1e4e8c10a396fae01/html5/thumbnails/20.jpg)
Source: https://towardsdatascience.com/battle-of-the-deep-learning-frameworks-part-i-cff0e3841750
![Page 21: Introduction to Amazon Sagemaker - Seoul AIseoulai.com › presentations › Introduction_to_Amazon_Sagemaker.pdf · Benefits of using machine learning in cloud 1. Size of dataset](https://reader030.vdocuments.mx/reader030/viewer/2022041022/5ed3bde1e4e8c10a396fae01/html5/thumbnails/21.jpg)
Conclusion
● Where is your data? - AWS vs GCP
● How big is your model? - Cluster vs Instance
● Try to use tf.estimator
● Which deep learning framework are you using?
○ Keep track model converters, model zoo
○ Training efficiency
○ Tutorials, examples, community
● Cost vs Time
![Page 22: Introduction to Amazon Sagemaker - Seoul AIseoulai.com › presentations › Introduction_to_Amazon_Sagemaker.pdf · Benefits of using machine learning in cloud 1. Size of dataset](https://reader030.vdocuments.mx/reader030/viewer/2022041022/5ed3bde1e4e8c10a396fae01/html5/thumbnails/22.jpg)
Thank you