Machine Learning at Netflix

Netflix’s ever changing libraries of movies, shows and original content have changed the way people watch TV. Netflix success is credited to pioneering ways that the company incorporated AI and Machine Learning (ML) into its products, services and infrastructure. Netflix is a data driven company. Data offers power insights for making informed business decisions. It fuels experimentation and innovation and helps Netflix discover content with a goal of predicting what our members want to watch next. To accelerate innovation velocity, Netflix has been building cloud native ML infrastructure that promotes: collaboration, model reuse and data abstraction.

Netflix notebook system and enhancements, ML libraries and orchestration help run hundreds and thousands of experiments every day. ML models should learn and adapt to new data as it arrives, and that is automated via CI/CD platform and DevOps support. Powerful multi-GPU cloud instances and distributed TensorFlow libraries making it possible to run more experiments and train more complex models in a short amount of time. At Netflix, ML is applied to tackle a wide range of challenges such as: personalizing member viewing experience, improving streaming quality, resource management, capacity forecasting, regional fail over and anomaly detection.