Learn about Google's powerful blackbox optimization and hyperparameter tuning system in this 42-minute AutoML seminar presentation. Dive into the technical architecture and capabilities of Open Source Vizier, a Python-based package that has optimized millions of machine learning models across Google's largest products. Explore the distributed system design, integration with Google's AutoML ecosystem, and the default Bayesian Optimization algorithm. Understand key technical challenges in distributed systems, scheduling mechanisms, data storage solutions, and various algorithm implementations. Discover how Vizier handles production-critical systems at scale, serving thousands of users reliably. Follow along as Richard Song breaks down the design choices, benefits, and integrations that make Vizier the go-to solution for experimental optimization and algorithmic benchmarking.
Open Source Vizier: Distributed Infrastructure for Blackbox Optimization and Hyperparameter Tuning
Overview
Syllabus
Introduction
Why a service
Scheduling jobs
Data Store
Algorithm
Default Algorithm
Open Source Algorithm
Design Choices
Benefits
Integrations
Taught by
AutoML Seminars
Related Courses
-
Hyperparameter Tuning with Keras Tuner
-
Cloud Computing and Distributed Systems
-
Hyperparameter Optimization and Algorithm Configuration from a Black-Box Optimization Perspective
-
Production Machine Learning Systems
-
Hyperparameter Tuning on the Cloud - Optimizing Machine Learning Models with SEER and Rubberband
-
Heterogeneity-Aware Algorithms for Federated Learning and Distributed Optimization
