Agent57: Outperforming the Atari Human Benchmark
DeepMind's Agent57 is the first RL agent to outperform humans in all 57 Atari benchmark games. It extends previous algorithms like Never Give Up and R2D2 by meta-learning the exploration-exploitation tradeoff controls.
https://arxiv.org/abs/2003.13350
https://deepmind.com/blog/article/Agent57-Outperforming-the-human-Atari-benchmark
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 propose an adaptive mechanism to choose which policy to prioritize throughout the training process. Additionally, we utilize a novel parameterization of the architecture that allows for more consistent and stable learning.
Authors: Adrià Puigdomènech Badia, Bilal Piot, Steven Kapturowski, Pablo Sprechmann, Alex Vitvitskyi, Daniel Guo, Charles Blundell
Links:
YouTube: https://www.youtube.com/c/yannickilcher
Twitter: https://twitter.com/ykilcher
BitChute: https://www.bitchute.com/channel/yannic-kilcher
Minds: https://www.minds.com/ykilcher
Видео Agent57: Outperforming the Atari Human Benchmark канала Yannic Kilcher
https://arxiv.org/abs/2003.13350
https://deepmind.com/blog/article/Agent57-Outperforming-the-human-Atari-benchmark
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 propose an adaptive mechanism to choose which policy to prioritize throughout the training process. Additionally, we utilize a novel parameterization of the architecture that allows for more consistent and stable learning.
Authors: Adrià Puigdomènech Badia, Bilal Piot, Steven Kapturowski, Pablo Sprechmann, Alex Vitvitskyi, Daniel Guo, Charles Blundell
Links:
YouTube: https://www.youtube.com/c/yannickilcher
Twitter: https://twitter.com/ykilcher
BitChute: https://www.bitchute.com/channel/yannic-kilcher
Minds: https://www.minds.com/ykilcher
Видео Agent57: Outperforming the Atari Human Benchmark канала Yannic Kilcher
Показать
Комментарии отсутствуют
Информация о видео
Другие видео канала
![](https://i.ytimg.com/vi/vJ9zTLZutno/default.jpg)
![Backpropagation and the brain](https://i.ytimg.com/vi/a0f07M2uj_A/default.jpg)
![DeepMind Made A Superhuman AI For 57 Atari Games! 🕹](https://i.ytimg.com/vi/dJ4rWhpAGFI/default.jpg)
![DeepMind's AlphaFold 2 Explained! AI Breakthrough in Protein Folding! What we know (& what we don't)](https://i.ytimg.com/vi/B9PL__gVxLI/default.jpg)
![Planning to Explore via Self-Supervised World Models (Paper Explained)](https://i.ytimg.com/vi/IiBFqnNu7A8/default.jpg)
![AlphaStar: Grandmaster level in StarCraft II using multi-agent reinforcement learning](https://i.ytimg.com/vi/BTLCdge7uSQ/default.jpg)
![DALL-E 2 by OpenAI is out! Live Reaction](https://i.ytimg.com/vi/gGPv_SYVDC8/default.jpg)
![AI learns to play pool](https://i.ytimg.com/vi/sx6eoXg-19k/default.jpg)
![LAION-5B: 5 billion image-text-pairs dataset (with the authors)](https://i.ytimg.com/vi/AIOE1l1W0Tw/default.jpg)
![Linformer: Self-Attention with Linear Complexity (Paper Explained)](https://i.ytimg.com/vi/-_2AF9Lhweo/default.jpg)
![Neural Architecture Search without Training (Paper Explained)](https://i.ytimg.com/vi/a6v92P0EbJc/default.jpg)
![AlphaGo - How AI mastered the hardest boardgame in history](https://i.ytimg.com/vi/MgowR4pq3e8/default.jpg)
![Improving Intrinsic Exploration with Language Abstractions (Machine Learning Paper Explained)](https://i.ytimg.com/vi/NeGJAUSQEJI/default.jpg)
![Discovering Symbolic Models from Deep Learning with Inductive Biases (Paper Explained)](https://i.ytimg.com/vi/LMb5tvW-UoQ/default.jpg)
![[Classic] Playing Atari with Deep Reinforcement Learning (Paper Explained)](https://i.ytimg.com/vi/rFwQDDbYTm4/default.jpg)
![Deep Q-Learning - Combining Neural Networks and Reinforcement Learning](https://i.ytimg.com/vi/wrBUkpiRvCA/default.jpg)
![Player of Games: All the games, one algorithm! (w/ author Martin Schmid)](https://i.ytimg.com/vi/U0mxx7AoNz0/default.jpg)
![Longformer: The Long-Document Transformer](https://i.ytimg.com/vi/_8KNb5iqblE/default.jpg)
![[ML News] GPT-3 learns to edit | Google Pathways | Make-A-Scene | CLIP meets GamePhysics | DouBlind](https://i.ytimg.com/vi/vGFaiLeoLWw/default.jpg)
![MuZero: DeepMind’s New AI Mastered More Than 50 Games](https://i.ytimg.com/vi/hYV4-m7_SK8/default.jpg)