There are applications of sim-to-real, e.g., in robotics. Why does it work and when will it fail?
3
Vote
L
S
Tony Qin
How do we handle the fact that the expected cumulative reward isn't usually the function you want to optimize for applications? I.e. you don't want a policy that works well 95% of the time.
1
Vote
Eugene
How should we best think about Type 1 error control wrt to learned policies?
by Matt Gershoff
For many real world problems, esp in marketing, we don't know apriori if any solution exists that...more
1
Vote
m
How to apply RL to recommendation and personalization product services? What are the pros and cons?
1
Vote
czx
Should we put more efforts in developing benchmarks closer to real life problems?
1
Vote
Noé
What are issues for RL in realistic applications?
1
Vote
S
What's your opinions on RL for science? For example, drug discovery, optimal design of biological sequences, etc.
0
Vote
WRT to optimizing utility functions (avoiding the issue of how to eliciting utility function from organizations) , I would be interested to know why aprior, we would expect RL to have at the margin greater efficacy to hand coding (domain knowledge) over the space of all decision/optimal control p...more
by Matt Gershoff
0
Vote
Are there realistic open datasets?
0
Vote
Can we do without models for safe and efficient RL in the real world?
