Securely executing Python machine learning models with distroless images at ING

Executing Python models safely and securely in production at ING

Thomas Kluiters

Data Science Docker Machine-Learning Operations Security

See in schedule Download Slides

Executing machine learning models in a production environment can be tricky, especially at a major bank where compliance and risk are carefully taken into account. In this talk I explain how, we, at ING (a large bank operating on global scale), execute our Python models in a production environment by building minimal Docker images for python versions.

I will first talk about the possible security risks of running any docker container in a production environment. Then I will talk about ways in which we can make Docker containers more secure by building minimal docker images for Python. Finally I will explain how these docker images are used in practice to serve machine learning models at ING.

Prerequisites:
- Some basic knowledge of Docker can be helpful
- Some basic understanding of security can be helpful

Goals:
- Understand the security risks of running docker containers
- Know how to make docker images more secure
- How to build secure model serving docker images

Type: Talk (30 mins); Python level: Beginner; Domain level: Beginner