Scaling ML Pipelines for Autonomous Driving with Ray
Explore how Ray revolutionizes machine learning pipelines for autonomous driving in this conference talk from Ray Summit 2024. Discover Motional's innovative approach to scaling critical workflows, including training data preparation and model evaluation. Learn how they achieved significant improvements, reducing data delivery latency from weeks to hours and attaining a 100X speedup in model evaluation. Gain insights into their custom metric computation library built on Ray Data, maximizing vectorization and enabling GPU deployment on heterogeneous Ray clusters. Delve into Motional's ML-based data curation solution, which dynamically selects valuable training data for each epoch, seamlessly integrating with their training process to potentially enhance model quality. Uncover practical applications of Ray in the autonomous driving industry and its transformative impact on large-scale machine learning workflows.