Ranjith Tharayil

Topic: Test driving Self learning systems to continuous delivery

Abstract: Test driving Machine Learning (ML) based software systems has always been a challenge to TDD and ATDD practitioners. In this talk, we will investigate one such approach that will make an AI software developer's job at ease by enabling her to drive the development guided by test. The key challenge for testing ML-based systems is the introduction of a new variable “the model” along with code. In conventional software development, the output of a system is always the function of the code, wherein in the case of an ML-based system the output of the system is a function of the code written and the model. This means the output of that system can change by only changing the model and models these days are designed to continuously evolve. This would mean traditional state testing methods will not work as the tests need to be modified every time the model changed. In this talk we will explore one such strategy that I experimented with and discovered during my experience; This 4 step strategy will enable developers to overcome the dilemma of how to test drive my ML-based software system.

Key takeaways
• How and Why testing (ML) based software systems is different from traditional software.
• Learn how the new 4 step approach will enable developers to test drive ML-based software system.
• The approach includes how to refactor and how-to strategies testing for various subsystems
• The participants should be able to apply the learning in their day to day work after the talk.