Introducing Amazon SageMaker JumpStart part 1 ML Solutions AWS re Invent 2020

Transcript

Hi, this is Julien from Arcee. In this video, I'd like to show you how you can easily deploy machine learning solutions using Amazon SageMaker Jumpstart, the new capability that we just launched. Jumpstart is integrated with SageMaker Studio, so that's where you should go first. Open the SageMaker Studio console, and you will see the "Explore 1.0 one-click solutions" box. Let's go to SageMaker Jumpstart and we can see a list of solutions. We can view all of those. Fraud detection, credit decisions, predictive maintenance, detecting malicious users, learning, demand forecasting, and more. Let's go all the way. Document understanding, product defect detection. That sounds cool. Let's do this one. This is about visual inspection, figuring out automatically if there's a defect, a manufacturing defect in a picture. We see some examples, and this is based on a reference data set. We can see the architecture used here and some extra links. Let's say this is a business problem that I care about, and I'd like to see how I can solve it using a machine learning model. Click on the launch button. This will run for a little while as SageMaker deploys the necessary AWS resources. This is based on AWS CloudFormation, our Infrastructure as Code service. If you go to the CloudFormation console, you'll see a stack being created with all the appropriate resources. However, we can happily ignore that fact and stay in SageMaker Studio and wait for this solution to be deployed. The solution has been deployed, and I see I can open a notebook for it. Clicking on this opens the first notebook because this solution actually has four notebooks. The first one deploys and predicts with a pre-trained model. The second notebook shows you how to do fine-tuning on the model. The third notebook shows you how to train the detector from scratch instead of fine-tuning. The last notebook shows you how to train a classifier from scratch. So, is there a defect or not in the image? You can see this is a detailed example. I suggest running the demo, understanding how this all works, and then trying it with your own images and your own data set. There's already a lot of existing code to start from. That's the whole point of ML solutions. You have an AWS architecture based on different services, and each solution will use a different set of services. You also have quite a few notebooks to understand the problem better and then customize it for your own data. So that's pretty cool. Have fun experimenting. When you're done, don't forget to delete the solution. This will delete all the AWS resources that were automatically created when we launched the solution. Don't forget to do that; otherwise, you're going to be charged. That's what I wanted to show you about ML solutions, end-to-end architectures, and examples to solve typical machine learning problems. Have fun experimenting. I'm happy to answer any questions you may have, and if you have any feedback, please feel free to get in touch. I'll see you soon with more videos. Bye.