+ Create new tricision Login 

Login

or


Forgot password?

First visit? Sign-up now!

By continuing you accept the Terms & Conditions and Privacy Policy.

time is up

RL + Operations Research Panel

Ask experts questions about RL + Operations Research. Panelists: Jim Dai (Cornell/CUHK-Shenzhen), Fei Fang (CMU), Shie Mannor (Technion & Nvidia Research), Yuandong Tian (Facebook AI Research); Co-Chairs: Zhiwei (Tony) Qin and Zongqing Lu (Moderator). Panel discussion time: 22:00-23:00, July 23 EDT (Boston time). https://sites.google.com/view/RL4RealLife


saved
Ideas
Pros and cons
 
Votes
What are some unique advantages that RL offers in solving OR problems compared to traditional methods like mathematical programming (which is by far still much more widely used)?
 
Deep RL algorithms solves scalability challenges in OR applications
by Jim
 
Deep RL can handle adaptivity quite well, especially when trained for that.
by Shie
 
While many OR problem instances are NP-hard (i.e., for any algorithm, there is always worst-case ... more
by Yuandong Tian
 
4
Vote

Fei, Mark, P and 1 more

Could each of you talk about one project at the interface of RL and OR that you are currently working on?
3
Vote
Jim
Mark
P
What are the most promising opportunities to combine RL with traditional OR methods, e.q., queueing theory, integer programming, to improve solutions?
2
Vote
Ahmad
Mark
How can we leverage the domain knowledge for training RL agents in this case? E.g. We know that (s,S) policy is the optimal form for stochastic lot-sizing problems under conditions.
2
Vote
Jim
Tes
How to make RL + OR work in real life? What are some unique issues that we have to deal with?
 
system load varies
by Jim
1
Vote
Jim
Which reinforcement libraries are apt for real life
1
Vote
Archana David
What are the most important and challenging RL problems in OR domain?
by Mark
1
Vote
Archana David
Which one do you think suits OR applications better? Model-based or model-free RL methods?
1
Vote
Mark
The field of OR is deeply grounded on theoretical foundations. On the other hand, the deep RL techniques emerged in recent years seem to be lacking in this aspect. Is this an issue that might hinder the adoption of RL for OR problems?
 
It is more of an opportunity than anything else
by Shie
1
Vote
Jim
What are the unique features of OR problems?
by Mark
0
Vote
Is there a good benchmark OR problem(s) to test/apply RL algorithms?
0
Vote
Is RL going to be one of the main stream approaches for OR?
0
Vote
What is the relationship between RL and OR?
0
Vote
From application perspective, what do you think are some promising future OR problems that are likely to benefit from RL?
0
Vote
Is sim2real as much an issue here as in other domains, say, robotics?
0
Vote
Wait, isn’t RL an OR problem itself?
0
Vote
What’s your advice to PhD students standing at this cross road?
0
Vote

Comments

https://www.tricider.com/brainstorming/3E5HWMMNSJp