AWS - MachineLearning

Posted on |

Hands-on

https://s3.amazonaws.com/solutions-reference/predictive-maintenance-using-machine-learning/latest/predictive-maintenance-using-machine-learning.pdf

1
Couldn't call 'describe_notebook_instance' to get the Role ARN of the instance PredictiveMaintenanceNotebookInstance.

Update the role attached to the sagemaker instance

1
ResourceLimitExceeded

Change to train_instance_type = ‘ml.p2.xlarge’

Reference

https://youtu.be/GW0Bktm55nI

AWS Machine Learning Stack

  • ML Frameworks & Infrastructures
    • Frameworks
      • Tensorflow ( 85% TensorFlow workloads in cloud runs on AWS)
      • Apache Mxnet – Deep learning for Enterprise dev ; liner scalable
      • Pytorch – Facebook ; flexible , versatile and portable
      • AWS is framework agnostic
  • ML Services
    • SageMaker workflows
    • SageMaker Ground Truth
    • Use SageMaker to do Re-enforced ML : For example Vehicle routing
    • Sagemaker Neo (Opensource)
      • Accelerate the cycle of doing Machine learning
      • CICD
      • Optimize between different frameworks
  • AI Services
    • Textract

GE Healthcare Demo

  • Neural Network Compression : reduce layer and retrain the model
  • How to archive network compression using AWS service
    • Use SageMaker RL
      • State current network archi
      • Action : remove layer or not
      • Reward : Accuracy + compression ratio
    • Result : 40% smaller model and 1%-2% loss of accuracy
Reward Makes Perfect
0%