Trends in Reinforcement Learning with Pablo Samuel Castro

Today we kick off our annual AI Rewind series joined by friend of the show Pablo Samuel Castro, a Staff Research Software Developer at Google Brain.

Subscribe: iTunes / Google Play / Spotify / RSS

Pablo joined us earlier this year for a discussion about Music & AI, and his Geometric Perspective on Reinforcement Learning, as well our RL office hours during the inaugural TWIMLfest. In today’s conversation, we explore some of the latest and greatest RL advancements coming out of the major conferences this year, broken down into a few major themes, Metrics/Representations, Understanding and Evaluating Deep Reinforcement Learning, and RL in the Real World. This was a very fun conversation, and we encourage you to check out all the great papers and other resources available below.

We want to hear from you! Send your thoughts on the year that was 2020 below in the comments, or via Twitter at @samcharrington or @twimlai.

To follow along with the 2020 AI Rewind Series, head over to the series page!

Thanks to our Sponsor!

Thanks to our friends at Pachyderm for sponsoring the show!

At the end of the day, real-world machine learning is all about the data. You already know this, but manually cleaning and transforming data can be exhausting, inconsistent, and error-prone, and is not the path towards getting your models into production, especially when your data, models, and code are constantly changing.

This is where Pachyderm can help. Pachyderm is an easy-to-use data science platform that lets you productionalize your machine learning tasks into fully-automated, end-to-end workflows, regardless of language or framework. Pachyderm provides Git-like data versioning and lineage that lets you automatically track every data change and final output result.

Right now, TWIML listeners can enjoy 20% off of Pachyderm Hub and build production-grade data science workflows in minutes, without ever having to configure a single piece of infrastructure.

Imagine being able to automate your entire data science workflow and reproduce any result from any point in time — in seconds, and with complete confidence.

Head over to pachyderm.com/TWIML to learn more and take advantage of this limited time offer.

Connect with Pablo

• Twitter
• Personal Page

Resources

• Music & AI Plus A Geometric Perspective on Reinforcement Learning
• Paper: Dynamical Distance Learning for Semi-Supervised and Unsupervised Skill Discovery
• Is a Good Representation Sufficient for Sample Efficient Reinforcement Learning?
• State Alignment-based Imitation Learning
• Contrastive Learning of Structured World Models
• CURL: Contrastive Unsupervised Representations for Reinforcement Learning
• Bootstrap Latent-Predictive Representations for Multitask Reinforcement Learning
• Planning to Explore via Self-Supervised World Models
• Representations for Stable Off-Policy Reinforcement Learning
• Dream to Control: Learning Behaviors by Latent Imagination
• Fast Task Inference with Variational Intrinsic Successor Features
• Model Based Reinforcement Learning for Atari
• Network Randomization: A Simple Technique for Generalization in Deep Reinforcement Learning
• Latent World Models For Intrinsically Motivated Exploration
• Measuring the Reliability of Reinforcement Learning Algorithms
• Revisiting Fundamentals of Experience Replay
• Behaviour Suite for Reinforcement Learning
• Explain Your Move: Understanding Agent Actions Using Specific and Relevant Feature Attribution
• Implementation Matters in Deep RL: A Case Study on PPO and TRPO
• What Can Learned Intrinsic Rewards Capture?
• Decentralized Reinforcement Learning: Global Decision-Making via Local Economic Transactions
• Munchausen Reinforcement Learning
• B-Suite
• Jax

Join Forces!

• Check out our TWIML Presents: series page!
• Register for the TWIML Newsletter
• Check out the official TWIMLcon:AI Platform video packages here!
• Download our latest eBook, The Definitive Guide to AI Platforms!

“More On That Later” by Lee Rosevere licensed under CC By 4.0