Google’s TensorFlow deep-learning API is a powerful tool that lets us take full advantage of the parallel processing capabilities offered by Graphical Processing Units (GPUs). With TensorFlow, we can train our neural networks faster, with greater control over our data processing pipeline.

This tutorial introduces TensorFlow through its low-level API. While it’s possible to jump straight to higher-level APIs like Keras which use TensorFlow as a backend, working first with the lower-level API gives us a better idea of what’s going on under the hood, which is useful for debugging and customization later on.

To demonstrate the use of TensorFlow, we’ll implement a convolutional neural network (CNN) that allows a computer to recognize handwritten digits. CNNs are a popular and powerful approach to many image-based classification tasks. 

Click to View Full Notebook

Notes:

• This IPython notebook is best viewed using Google Chrome; some images and hyperlinks may not work in Mozilla FireFox.
• To download the source code, which you can edit and execute yourself, save this link (.ipynb file extension).