"Digital ad campaign optimization engines need to obtain machine learning predictions within a few milliseconds."Sergei Izrailev Chief Data Scientist
- Advertising Optimization
- Design Patterns
Overview of the Challenge
Real-time systems such as those deployed in online advertising impose stringent constraints on the design of machine learning solutions. Digital ad campaign optimization engines typically need to obtain machine learning predictions for a given ad within a few milliseconds. At the same time, the number of ads ranges from tens of thousands to millions per second. This presentation describes general design patterns that tend to work well in such an environment and illustrates how real-time constraints shaped machine learning system design decisions at Beeswax.