Melanoma Prediction Using CNN Methods. Part 7

Part 6 ML Software Component Preparation

ML Model Serving Service Deployment Options

When I have trained machine learning model, usually I have two options for using the model in my product. One of the options is embedded model into the iOS application. And another option is providing API for connection to the deployed ML model serving service, which provides functionality of loading the trained model.

There are many options as AWS services (AWS EC2, 2022), GCP services (Google vertex ai, 2022), Azure services (docs.microsoft.com, machine learning, 2022), Heroku cloud provider (Heroku, 2022).

On the one hand, most of the cloud providers suggested of using serverless services as AWS Lambda or GCP Vertex AI. The services provide automated approach of using AI features.

On the other hand, cloud providers have flexibility for clients, and they suggest only machines with operation system, even with opportunity to choose the operation system. For instance, AWS cloud provider has EC2 instances.

With reference to my time management, I reviewed three cloud providers. Table 1 contains cloud providers, which I investigated and reviewed.

In scope of my project, I investigated the Heroku cloud platform with docker deployment as one of cheap options, GCP Vertex AI serverless service, and I reviewed and finally chose to use Amazon Web Services.

References

AWS EC2, 2022, Amazon Elastic Compute Cloud (Amazon EC2). Available at: https://aws.amazon.com/ec2/

[Accessed 04 July 2022]

Google vertex ai, 2022, Vertex AI. Available at: https://cloud.google.com/vertex-ai

[Accessed 20 May 2022]

docs.microsoft.com, machine learning, 2022, What is Azure Machine Learning? Available at: https://docs.microsoft.com/en-us/azure/machine-learning/overview-what-is-azure-machine-learning

[Accessed 14 July 2022]