Introduction Today topic is about continuous model deployment. When working with Machine Learning systems, one must always remember that the world around is changing fast. Training a model, deploying it online and calling it a day is not a valid option, as model's performance degrades quickly. If you want to...
Introduction In this article, I will describe how Yelp predicts the waiting time for restaurants around the world. In this setting, latency of the system is paramount: when users want to know the current waiting time, they expect an immediate answer. However, they don't particularly care for the system to...
Introduction The harsh reality of new graduates joining a Machine learning team is that the "model development" part of the job dwarfs in comparison to everything else the team needs to support. The priority of the team is often: improving models' up-time, optimizing the data pipeline, adding new monitoring dimensions...
Introduction In today's article, I am going to dive deep into the paper [1] published by TikTok engineers describing how TikTok real time recommendation system is built. The paper has many topics that are worth learning in more detail, but I am going to focus on: * How the design supports...
Hey! I am Ludo. I am a machine learning engineer at Google. I have created "Machine learning at scale" to talk about ML systems in the real world. I find that the majority of online ML content is divided in these two groups: * Introduction to model X in Python. * An...
