#AI and #Neuroscience: A virtuous circle via #DeepMindAI https://buff.ly/2feh4DQ https://t.me/ArtificialIntelligenceArticles
A wonderful comprehensive read from #Google_Brain and #DeepmindAI on the challenges which we can come across while implementing RL on real-world systems.
Paper-Title: Challenges of Real-World Reinforcement learning
Link to the paper: https://arxiv.org/abs/1904.12901
They highlighted 9 most important challenges as follows:
1. Training off-line from the fixed logs of an external behavior policy.
2. Learning on the real system from limited samples.
3. High-dimensional continuous state and action spaces.
4. Safety constraints that should never or at least rarely be violated.
5. Tasks that may be partially observable, alternatively viewed as non-stationary or stochastic.
6. Reward functions that are unspecified, multi-objective,or risk-sensitive.
7. System operators who desire explainable policies and actions.
8. Inference that must happen in real-time at the controlfrequency of the system.
9. Large and/or unknown delays in the system actuators,sensors, or rewards.
Paper-Title: Challenges of Real-World Reinforcement learning
Link to the paper: https://arxiv.org/abs/1904.12901
They highlighted 9 most important challenges as follows:
1. Training off-line from the fixed logs of an external behavior policy.
2. Learning on the real system from limited samples.
3. High-dimensional continuous state and action spaces.
4. Safety constraints that should never or at least rarely be violated.
5. Tasks that may be partially observable, alternatively viewed as non-stationary or stochastic.
6. Reward functions that are unspecified, multi-objective,or risk-sensitive.
7. System operators who desire explainable policies and actions.
8. Inference that must happen in real-time at the controlfrequency of the system.
9. Large and/or unknown delays in the system actuators,sensors, or rewards.
arXiv.org
Challenges of Real-World Reinforcement Learning
Reinforcement learning (RL) has proven its worth in a series of artificial domains, and is beginning to show some successes in real-world scenarios. However, much of the research advances in RL...
Paper-Title: COBRA: Data-Efficient Model-Based RL through Unsupervised Object Discovery and Curiosity Driven Exploration
#DeepmindAI
Link to the paper: https://arxiv.org/pdf/1905.09275.pdf
The three specific tasks performed in this paper using SpriteWorld(https://github.com/deepmind/spriteworld) are:-
Goal-finding task. The agent must bring the target sprites (squares) to the centre of the arena.
Clustering task. The agent must arrange the sprites into clusters according to their colour.
Sorting task. The agent must sort the sprites into goal locations according to their colour (each colour is associated with a different goal location).
The main technical contributions of the paper are:-
A method for learning action-conditioned dynamics over slot-structured object-centric representations that require no supervision and is trained from raw pixels.
A method for learning a distribution over a multi-dimensional continuous action space. This learned distribution can be sampled efficiently.
An integrated continuous control agent architecture that combines unsupervised learning, adversarial learning through exploration, and model-based RL.
#DeepmindAI
Link to the paper: https://arxiv.org/pdf/1905.09275.pdf
The three specific tasks performed in this paper using SpriteWorld(https://github.com/deepmind/spriteworld) are:-
Goal-finding task. The agent must bring the target sprites (squares) to the centre of the arena.
Clustering task. The agent must arrange the sprites into clusters according to their colour.
Sorting task. The agent must sort the sprites into goal locations according to their colour (each colour is associated with a different goal location).
The main technical contributions of the paper are:-
A method for learning action-conditioned dynamics over slot-structured object-centric representations that require no supervision and is trained from raw pixels.
A method for learning a distribution over a multi-dimensional continuous action space. This learned distribution can be sampled efficiently.
An integrated continuous control agent architecture that combines unsupervised learning, adversarial learning through exploration, and model-based RL.
GitHub
deepmind/spriteworld
Spriteworld: a flexible, configurable python-based reinforcement learning environment - deepmind/spriteworld