PyCon X

Florence

2nd - 5th May 2019

Everything your model doesn't know

So you trained your machine learning model. The accuracy is very good, and it’s now time for deployment in production. And you start to wonder about the hundreds of ways in which a data-driven black box can go awfully wrong: bias in the data, adversarial attacks on your model, drifting features, strange inputs crashing your entire system… You are not feeling so over-confident anymore!

In this talk I will provide a brief overview of recent (and not-so-recent) ideas and tools for mitigating these problems. We will see some interesting Python libraries for analyzing more in-depth your model and checking its reliability over a wide range of practical issues. Only an introductory knowledge in machine learning and Python is required to fully appreciate the talk.

Feedback form: https://python.it/feedback-1524


Do you have some questions on this talk?

New comment