Agent57: Outperforming the Atari Human Benchmark

First Virtual London Machine Learning Meetup - 13.05.20 @ 18:30

After a short pause, we would like to invite you to our first virtual machine learning meetup. We are taking the opportunity to change the format slightly and devote more time to Q&A. Please read the papers below and help us create a vibrant discussion.

The discussion will be facilitated by Sam Ritter, research scientist at DeepMind.

Login in details will be announced soon.

Agenda:

• 18:25: Virtual doors open
• 18:30: Talk
• 19:00: Q&A session
• 19:35: Close

Sponsors
Man AHL: At Man AHL, we mix machine learning, computer science and engineering with terabytes of data to invest billions of dollars every day.

Evolution AI: Machines that Read - get answers from your text data.

• Title: Agent57: Outperforming the Atari Human Benchmark (Charles Blundell - DeepMind, Gatsby UCL)
• Paper:
  https://arxiv.org/abs/2003.13350
  https://arxiv.org/abs/2002.06038

Abstract: Atari games have been a long-standing benchmark in the reinforcement learning (RL) community for the past decade. This benchmark was proposed to test general competency of RL algorithms. Previous work has achieved good average performance by doing outstandingly well on many games of the set, but very poorly in several of the most challenging games. We propose Agent57, the first deep RL agent that outperforms the standard human benchmark on all 57 Atari games. To achieve this result, we train a neural network which parameterizes a family of policies ranging from very exploratory to purely exploitative. We shall review what goes into this agent, and how it achieves this level of performance.

Bio: Charles Blundell is a research scientist at DeepMind where he leads a team who have recently been working on exploration in reinforcement learning and combining episodic memory with deep learning. He holds a PhD in machine learning from the Gatsby Unit at UCL.