Explore TF-Agents, the latest reinforcement learning library for TensorFlow, in this 50-minute presentation by Software Engineers Oscar Ramirez and Sergio Guadarrama. Discover why supervised learning is insufficient for certain tasks and learn about the key differences between supervised and reinforcement learning. Gain insights into measuring the reliability of RL algorithms and see a demonstration of playing Breakout faster using TensorFlow. Delve into topics such as learnable policies, policy savers, and available agents. Understand the distributed collection and architecture in TF-Agents, and explore the differences between bandits and reinforcement learning. Learn about A/B testing, multi-armed bandits, and their applications in recommender systems. Finally, get a glimpse of the TF-Agents roadmap and the developer investment behind this powerful library.
Overview
Syllabus
Intro
Supervised learning is not enough
Differences with supervised learning
Typical supervised learning
Typical reinforcement learning
Measuring the reliability of RL algorithms
Playing Breakout faster in TF
Learnable policy
Policy Saver
Available agents
TF-Agents distributed collection
Distributed architecture
Bandits vs RL
A/B testing
Multi-armed bandits
Bandit agents
Recommender systems
TF-Agents: Roadmap
TF-Agents developer investment
Taught by
TensorFlow
