In this chapter, we've gone through an overview of convolutional neural and residual networks. In addition, we've illustrated how SageMaker's image classification algorithm can be used to identify fast-food and bakery images. Specifically, we've reviewed training an image classification algorithm, including provisioning its infrastructure; creating a compressed image format (RecordIO) for training and validation datasets; and supplying formatted datasets for model fitting. For inference, we've employed the Batch Transform feature of SageMaker to classify multiple images in one go.
Most importantly, we've learned how to apply transfer learning to image classification. This technique becomes very powerful in instances where you do not have large amounts of training data.
In the next chapter, you'll learn how to forecast retail sales using...
