+ 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 Foundation Panel

Ask experts questions about the RL Foundation. Moderator: John Langford (Microsoft). Panelists: Matthew Botvinick (Deepmind), Thomas Dietterich (Oregon State U.), Leslie Pack Kaelbling (MIT), Warren Powell (Princeton & Optimal Dynamics). Co-Chairs: Csaba Szepesvari, Lihong Li and Yuxi Li. Panel discussion time: 11:00-12:00, July 23 EDT (Boston time). https://sites.google.com/view/RL4RealLife


saved
Ideas
Pros and cons
 
Votes
Is reward enough? C.f. “Hypothesis (Reward-is-Enough). Intelligence, and its associated abilities, can be understood as subserving the maximisation of reward by an agent acting in its environment.” David Silver, Satinder Singh, Doina Precup, Richard S. Sutton, Reward is enough. Artificial Intelli... more
 
4
Vote

Ahmad Ridley, Yashvir, Frederik Schubert and 1 more

What is the relationship between RL, theoretical CS/ML, classical AI, neuroscience, operations research, optimal control, stochastic optimization? Is it possible to build synergy among these disciplines? How?
2
Vote
Ahmad Ridley
Michal Lisicki
What is the foundation for RL?
 
Knowledge Representations
 
States and its categorisations
1
Vote
Y
What will be the impact of hybrid AI approaches, such as neuro-symbolic AI, on the future of real-world RL?
 
Pro: Addition of symbolic, logical reasoning would presumably be required to extend performance o... more
 
Con: Addition of inductive bias of expert knowledge might not be necessary and negatively impact ... more
1
Vote
Ahmad Ridley
Are you aware of any good problem examples for testing/benchmarking theoretical ideas for RL algorithms like Logistic regression in convex optimization?
by Eg
1
Vote
S
How does the RL foundation influence/inform RL for Real Life? What insights RL practitioners may learn/borrow from the RL foundation?
1
Vote
-
What will be the approach for RL to be successful, in theory and in practice, general methods like search and meta-learning with sufficient compute, inductive bias, engineering? (Rich Sutton’s Bitter lesson; Rodney Brooks’ Better lesson; Leslie Pack Kaelbling’s Engineering AI)
1
Vote
Frederik Schubert
Representations
 
Invariance, Equivariance, Symmetry, etc Hierarchical latent spaces
0
Vote
Evolutionary Strategy (E.S) vs RL
0
Vote
Where do you see RL in 20 years? What further milestones of the RL field, in general, do you expect (e.g. beating humans in Go)? To what other domains can RL potentially contribute or considerably increase its impact?
by Danil
0
Vote
Should we consider solving POMDP rather than developing algorithm in MDP setting?
by R
0
Vote
What do each of the panelists think are the biggest open problems in RL today that get in the way of broader adoption of RL in production
by Nikhil
 
Robustness Safety Scalability Exploration vs Exploitation
0
Vote

Comments

https://www.tricider.com/brainstorming/2sca6QniPtp