Description:

The Machines are on the rise! Everybody is aware of the ever-growing presence of machine learning algorithms all around us. The Machines will soon rule the world. At least that’s what the vendors try to tell us.

Every piece of software that claims to provide a solution, should also be testable to prove that it actually is a solution for the given problem. When it comes to testing of machine learning the excuse is often: "We cannot test this, it’s ML!“ It’s time to do something against that!

Not too many have taken the time yet to understand what machine learning actually does, and how to test these algorithms?

In this talk I want to give the audience a chance to make the first step and understand the very basics about Machine Learning algorithms based on the "Hello World“ example of Machine Learning. The handwritten digit recognition algorithm for the MNIST database. 

Every tester needs a good model to understand what to test. I will explain the basics of the mystical tensor layers, and we take a glimpse into them so that you get an idea of what actually happens in there.

We will touch the topics of how testing works in this and some other examples and how observability and traceability look like. And we will take a look at a few recognition failures to understand why machine learning is not a black and white pass/fail scenario anymore and why some of the approaches to testing that we used for decades start to fail.