Meta Leverages Machine Learning to Optimize Network Performance Across Apps
-
Adopted a machine learning approach to optimize bandwidth estimation and congestion control across Meta's apps
-
Use time series data and simulations to categorize network types and tune parameters for each type
-
Developed models to classify packet loss and predict congestion before it happens
-
Experiment results showed wins for reliability, quality, and engagement across various network conditions
-
Future plans include consolidating models, improving simulations, and using ML to recommend optimal network actions