2019-06-03, 15:30–15:55, Hall 3
A practical guide for defining effective interfaces between Python applications and Python-based Deep Learning algorithms
Update: Slides available at http://bit.ly/317Dpqe
One untalked merit of the Deep Learning revolution is the dramatic change in how we build Software; the old days of a developer re-implementing the algorithmic sketch are long gone, as frameworks like TensorFlow reduce the distance from a theoretical paper to production to nearly zero.
But is it without a price? What happens when the requirements changes? How can new features be added without requiring to train a new model? How can we continuously improve our DL algorithms without requiring SW changes?
In this talk I’ll present a practical interface which allows continuous development of both your Python application, and your Deep Learning models. We’ll walk through the different trade-offs, and how changes can be introduced on both sides, while keeping your system live in production.