Given the limitations of interacting on-line with many real-world systems that could benefit from RL, should there be more emphasis on safe batch RL methods?
Recently, there has been a number of attempts to formulate realistic RL benchmarks / environments (PARK, Realistic Half Cheetah, NAS-Bench, Internet Congestion Control, SmartChoices to name a few) but no agreed structure of environment yet. However, is the notion of traditional "environment" real...more
by Andrii Maksai
4
Vote
Thomas Deselaers, Taylor Killian, Hyunseok Kim and 1 more
What applications are not suitable for (the current) RL? (Why should we not choose RL?)
How important is using simulation for RL? Can we do without simulators?
by Blazej
3
Vote
Blazej
M
Steven
In order to solve more challenging problems, do we need solutions which are generic and focus on meta-methods, or should we focus on incorporation priors?
3
Vote
Matthias
S
Steven Latre
What are the common pitfalls of applying RL to the real world problems?
3
Vote
A
B
Xian Yeow Lee
How does Quantum computing effect Reinforcement Learning?
by Kyle Saltmarsh
2
Vote
-
Jack
David Silver: How to understand RL + DL = AI?
2
Vote
Anna_Luo_Amazon
RLRL
How do we explore vs. exploit in the real-world?
2
Vote
Andrii
Shaswot
How is RL related to optimal control? How about RL vs. approximate dynamic programming (ADP)? Any advices if using RL for traditional optimal control and ADP applications?
2
Vote
Chun
Thomas Deselaers
John mentioned that we should build foundations while also reaching for the stars. What exactly are the places where our RL foundations need reinforcement (no pun intended)?
by Shaswot
1
Vote
doubling
Can RL ideas in their current state help individuals in the world on a personal level?
by Evan Fuller
Meaning without a corporate or academic development process. E. G. Could one deploy a simple tool...more
1
Vote
Evan
The paper "Quantifying Generalization in Reinforcement Learning" indicates that most RL research at the moment doesn't generalize at all. Does that mean that most of the recent improvements reported on Atari, Mujoco etc. will not easily be reproducible in the real world?
by Thomas Deselaers
1
Vote
Ian Quah
What techniques (meta, hierarchical, ...) do you consider key to make RL more sample efficient?
1
Vote
B
RL addresses the very important but unwieldy problem of artificial intelligence. Can we compartmentalize some of RL's success to alleviate issues that supervised learning alone cannot solve?
1
Vote
Carl
How much of the real world to learned value functions reflect? Do we expect them to?
1
Vote
Taylor Killian
Deep RL is very popular and powerful in current research. Is non-deep RL still useful in practice? Why?
1
Vote
Thomas Deselaers
How do I convince my CEO/CTO to try RL on our products/services? (Why should we choose RL?)
1
Vote
Alex
Sim-to-real sees some applications, e.g., in robotics. Why does it work and when will it fail? Are there theoretical guidelines?
1
Vote
Hyunseok Kim
To make RL for real life applications successful, do we need to think alternatives to RL? What are alternatives to RL? Do we treat imitation learning, game theory, causal inference as alternatives or complementaries to RL?
1
Vote
Rob
What are good books about RL?
What's your background in it? Hard to give good advice not knowing what you know
0
Vote
Personalized healthcare from RL - what's needed to get there ?
Is causality the roadblock ?
Doc.Ai already solved it, didn't you hear?
by -
0
Vote
How promising are evolutionary strategies as a scalable alternative to RL?
0
Vote
When is AlphaStar going into Beta?
0
Vote
Should we put more research efforts in development benchmarks closer to real-world problems?
0
Vote
What are the challenges for safety-critical RL system like autonomous cars and autonomous aircraft?
by Peng Wei
0
Vote
Are there any "live/real" use of off-policy RL in healthcare for treatment optimization ? Was(is) IBM Watson very much based on this ?
0
Vote
Should we consider/discuss approaches that make more sense if the agent is likely to find itself in a null world?
by Matt Gershoff
To elaborate, if one aspect of real world RL is to put RL in the hands of business users/marketer...more
0
Vote
What approach/algorithms would you recommend for recommendation and personalization product services to start with if they are interested in exploring the reinforcement learning approach? Can you please elaborate the pros and cons of the approach and how long it might take?
