4-2Flink在AI流程中的應用.pdf

1、Background on Flink with AI What is AI flow Meet Flink AI flow ML scenario in Flink AI flow Background on Flink with AI Lambda architecture Data Source Batch Processing Stream Processing Merge the results Batch layer Speed layer Offline path Online path Downstream Analytics Serving layer AI requirem

2、ents for real-time property Data Preprocessing big data processing essentially maintain two sets of codes Training sample distribution deviation, may need train model regularly high frequency search scenario require online training Predict offline , online, nearline all require lowest latency Why Fl

3、ink? Offline path Online path Data Source ProcessingOffline TrainingInference Processing Static Model Message Queue Online TrainingInference Processing Dynamic Model Why Flink? Data SourceProcessingOffline Training Message Queue Online Training Inference Processing Dynamic Model Static Model Offline path Online path Static Model What is AI flow An library managing the machine learning pipeline lif