Building products with TensorFlow
What makes a great product? At Google we call it the ‘magic moments’ – when your email suggests an insightful automatic response, when the map knows to highlight restaurants as you drive by hungry or when the ad campaign you are building offers to change its parameters to maximize what your brand cares about. Or when you don’t notice anything at all but the fraudulent transaction did not impact your account or that spam message did not hit your mailbox.
But how does one get there? How to tell if a problem is suitable for ML? How much data is actually needed? Should one build custom models or just use the existing services like Google’s Cloud ML? How expensive will the whole solution be and what kind of wins can one expect?
Armed with a few rules of thumb on how to make the right call, we will look at various business problems and their example solutions ranging from a few ML API calls for standard tasks to full fledged TensorFlow solutions where we cover:
- Data pre-processing and encoding
- Network topology & learning: LSTM, GANs, VAEs
- Visualization and evaluation using TensorBoard