Difference between revisions of "Deep Reinforcement Learning (DRL)"
m |
m |
||
| Line 58: | Line 58: | ||
* [http://deepmind.com/blog/impala-scalable-distributed-deeprl-dmlab-30/ Importance Weighted Actor-Learner Architectures: Scalable Distributed DeepRL in DMLab-30] | * [http://deepmind.com/blog/impala-scalable-distributed-deeprl-dmlab-30/ Importance Weighted Actor-Learner Architectures: Scalable Distributed DeepRL in DMLab-30] | ||
| − | uses resources more efficiently in single-machine training but also scales to thousands of machines without sacrificing data efficiency or resource utilisation. We achieve stable learning at high throughput by combining decoupled acting and learning with a novel off-policy correction method called V-trace. IMPALA is able to achieve better performance than previous agents with less data, and crucially exhibits positive transfer between tasks as a result of its multi-task approach. | + | uses resources more efficiently in single-machine training but also scales to thousands of machines without sacrificing data efficiency or resource utilisation. We achieve stable learning at high throughput by combining decoupled acting and learning with a novel off-policy correction method called V-trace. IMPALA is able to achieve better performance than previous [[agents]] with less data, and crucially exhibits positive transfer between tasks as a result of its multi-task approach. |
http://ai2-s2-public.s3.amazonaws.com/figures/2017-08-08/cbddc41e5b07ad8e3f7982e232bafba84c8419cc/5-Figure3-1.png | http://ai2-s2-public.s3.amazonaws.com/figures/2017-08-08/cbddc41e5b07ad8e3f7982e232bafba84c8419cc/5-Figure3-1.png | ||
| Line 74: | Line 74: | ||
* [http://github.com/deepmind/lab/tree/master/game_scripts/levels/contributed/dmlab30 DMLab-30 | GitHub] | * [http://github.com/deepmind/lab/tree/master/game_scripts/levels/contributed/dmlab30 DMLab-30 | GitHub] | ||
| − | * [http://www.techleer.com/articles/488-impala-distributed-agent-in-dmlab-30/ IMPALA distributed agent in DMLab-30 | + | * [http://www.techleer.com/articles/488-impala-distributed-agent-in-dmlab-30/ IMPALA distributed] [[Agents|agent]] in DMLab-30 |
* [http://www.semanticscholar.org/paper/IMPALA%3A-Scalable-Distributed-Deep-RL-with-Weighted-Espeholt-Soyer/cbddc41e5b07ad8e3f7982e232bafba84c8419cc IMPALA: Scalable Distributed Deep-RL with Importance Weighted Actor-Learner Architectures] | * [http://www.semanticscholar.org/paper/IMPALA%3A-Scalable-Distributed-Deep-RL-with-Weighted-Espeholt-Soyer/cbddc41e5b07ad8e3f7982e232bafba84c8419cc IMPALA: Scalable Distributed Deep-RL with Importance Weighted Actor-Learner Architectures] | ||
http://s3.ap-south-1.amazonaws.com/techleerimages/4d62b60c-4dcd-4774-9c75-417eba1cbbc1.png | http://s3.ap-south-1.amazonaws.com/techleerimages/4d62b60c-4dcd-4774-9c75-417eba1cbbc1.png | ||
Revision as of 08:01, 4 February 2023
Youtube search... ...Google search
- Reinforcement Learning (RL)
- Monte Carlo (MC) Method - Model Free Reinforcement Learning
- Markov Decision Process (MDP)
- State-Action-Reward-State-Action (SARSA)
- Q Learning
- Deep Reinforcement Learning (DRL) DeepRL
- Distributed Deep Reinforcement Learning (DDRL)
- Evolutionary Computation / Genetic Algorithms
- Actor Critic
- Hierarchical Reinforcement Learning (HRL)
- MERLIN: Inside Out - Curious Optimistic Reasoning
- OpenAI Gym
OTHER: Policy Gradient Methods
_______________________________________________________________________________________
- Introduction to Various Reinforcement Learning Algorithms. Part I (Q-Learning, SARSA, DQN, DDPG) | Steeve Huang
- Introduction to Various Reinforcement Learning Algorithms. Part II (TRPO, PPO) | Steeve Huang
- Guide
Goal-oriented algorithms, which learn how to attain a complex objective (goal) or maximize along a particular dimension over many steps; for example, maximize the points won in a game over many moves. Reinforcement learning solves the difficult problem of correlating immediate actions with the delayed returns they produce. Like humans, reinforcement learning algorithms sometimes have to wait a while to see the fruit of their decisions. They operate in a delayed return environment, where it can be difficult to understand which action leads to which outcome over many time steps.
Importance Weighted Actor-Learner Architecture (IMPALA)
YouTube search... ...Google search
- Artificial General Intelligence Is Here, and Impala Is Its Name | Aaron Krumins
- DeepMind Lab
- IMPALA: Scalable Distributed Deep-RL with Importance Weighted Actor-Learner Architectures
- Importance Weighted Actor-Learner Architectures: Scalable Distributed DeepRL in DMLab-30
uses resources more efficiently in single-machine training but also scales to thousands of machines without sacrificing data efficiency or resource utilisation. We achieve stable learning at high throughput by combining decoupled acting and learning with a novel off-policy correction method called V-trace. IMPALA is able to achieve better performance than previous agents with less data, and crucially exhibits positive transfer between tasks as a result of its multi-task approach.
DMLab-30
Youtube search... ...Google search
DMLab-30 is a collection of new levels designed using our open source RL environment DeepMind Lab. These environments enable any DeepRL researcher to test systems on a large spectrum of interesting tasks either individually or in a multi-task setting.
- DMLab-30 | GitHub
- IMPALA distributed agent in DMLab-30
- IMPALA: Scalable Distributed Deep-RL with Importance Weighted Actor-Learner Architectures
