报告题目：End-to-end Optimization of Machine Learning Prediction Queries

论文出处：SIGMOD 2022

作者：Kwanghyun Park, Karla Saur, Dalitso Banda

单位：Microsoft

报告人：胡华彦

报告时间：2022年11月14日 下午 1:00

报告地点：贵州大学北校区博学楼624室

报告内容摘要：Prediction queries are widely used across industries to performadvanced analytics and draw insights from data. They include adata processing part (e.g., for joining, filtering, cleaning, featurizingthe datasets) and a machine learning (ML) part invoking one ormore trained models to perform predictions. These parts have sofar been optimized in isolation, leaving significant opportunitiesfor optimization unexplored. We present Raven, a production-ready system for optimizingprediction queries. Raven follows the enterprise architectural trendof collocating data and ML runtimes. It relies on a unified interme-diate representation that captures both data and ML operators in asingle graph structure to unlock two families of optimizations. First, it employs logical optimizations that pass information between the data part (and the properties of the underlying data) and the ML part to optimize each other. Second, it introduces logical-to-physical transformations that allow operators to be executed on different runtimes (relational, ML, and DNN) and hardware (CPU, GPU). Novel data-driven optimizations determine the runtime to beused for each part of the query to achieve optimal performance. Ourevaluation shows that Raven improves performance of predictionqueries on Apache Spark and SQL Server by up to 13.1× and 330×, respectively. For complex models where GPU acceleration is benefi-cial, Raven provides up to 8× speedup compared to state-of-the-art systems.