In November 2015, Google released an open source software platform for deep learning called TensorFlow. TensorFlow has a flexible architecture, which allows it to spread computations across multiple CPUs or use a Graphical Processing Unit (GPU). GPUs have thousands of computing cores, which facilitate massive parallel processing and align well with deep learning training needs. The most common way to interact with it is with Python code.

The Python keras package acts as an interface layer to TensorFlow, making programming simpler. The author recommends using keras instead of programming TensorFlow directly.

So, how can you easily set up TensorFlow with the keras interface all on top of a GPU unit to accelerate training times? Fortunately, AWS provides AMI instances with all of this, and more, already included. You can easily launch one onto an EC2 instance with GPU support (make sure to check pricing first - GPUs are in high demand). These are called AWS Deep Learning AMIs, and information about how to get started can be found at https://aws.amazon.com/amazon-ai/amis/.