Overview
Explore adaptive experimentation techniques for product optimization in this 21-minute conference talk from Facebook's F8 2019 event. Learn how Facebook AI applies weakly supervised learning and multimodal understanding at scale to improve computer vision research. Discover the process of raising awareness for blood donation features and the shift from static to personalized configurations. Examine Bayesian models' ability to capture possible realities and their application in optimizing HHVM's Just-In-Time Compiler. Understand how Ax empowers developers to tune configurations and the practical implementation of adaptive experimentation. Gain insights into the adaptive experimentation ecosystem and the flexible research platform Bo Torch for Bayesian optimization.
Syllabus
Optimizing Products with Adaptive Experimentation
Raising Awareness of Blood Donation Features
The Status Quo: Static Configuration
Personalized Configuration
Many Models Are Compatible with the Data
Bayesian Models Capture Possible Realities
Optimizing HHVM's Just-In-Time Compiler
Ax Empowers Developers to Tune Configurations
Adaptive Experimentation Ecosystem
Adaptive Experimentation in Practice
Bo Torch A Flexible Research Platform for Bayesian Optimization
Taught by
Meta Developers