A team using Amazon SageMaker finds the default built-in image classification algorithm (ResNet) gives low accuracy and they want to use an Inception architecture instead. Which approaches will let them train with Inception? (Choose two.)

The two correct approaches leverage SageMaker's flexibility for custom models. Providing a Docker container with a TensorFlow Estimator and Inception network allows you to package your specific environment and model architecture, which SageMaker then uses for training. Similarly, writing custom training code directly in SageMaker, utilizing a TensorFlow Estimator and loading an Inception network, gives you direct control over the model implementation within SageMaker's managed environment. Modifying a built-in algorithm is not possible as they are pre-configured. Opening a support case for a custom algorithm change is not how SageMaker operates; users are expected to bring their own models for specific needs. Installing Inception on an EC2 instance and using it as a Jupyter notebook within SageMaker is inefficient and doesn't utilize SageMaker's managed training capabilities; it's essentially running a local notebook on a remote server.