Follow Us On

AutoML by-and-for Reinforcement Learning

January 14, 202016:00 - 16:30State of AI and ML - January 2020

Speaker

Yifang Liu

Yifang Liu

VP of Machine Learning Smule Inc.
  • January 14, 2020
  • 04:00pm - 04:30pm

Abstract

Reinforcement learning is a very powerful tool for important machine learning problems across industries, especially for those related to behavior modeling. Meanwhile, as the currently manual tasks of hyperparameter tuning and feature selection become increasingly inefficient due to the more and more complex models, AutoML is emerging as a highly-desired machine learning technology to take on the automation of those insight-required tasks. Today, reinforcement learning algorithms still face the tough challenge of hyperparameter optimization, such as how to find the best representation of individual states and the most effective rewards. This talk gives a brief overview of the state-of-the-art methods to cope with the challenge and proposes a novel AutoML approach that is based on reinforcement learning, and that can be applied to reinforcement learning and other ML algorithms.