+ 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 Explainability & Interpretability Panel

Ask experts questions about RL Explainability & Interpretability. Panelists: Ofra Amir (Technion), Finale Doshi-Velez (Harvard), Alan Fern (Oregon State U.), Zachary C. Lipton (CMU); Co-Chairs/Moderators: Omer Gottesman and Niranjani Prasad. Panel discussion time: 12:00-13:00, July 23 EDT (Boston time). https://sites.google.com/view/RL4RealLife


saved
Ideas
Pros and cons
 
Votes
What's the representations that is recommended for building an RL framework from scratch? Is there some standard way or guidelines which a practitioner needs to follow for plugging into an explanable tool? Eg for dealing with representing policies amongst other things.
by Mayank Bhaskar
 
Eg: https://distill.pub/2020/understanding-rl-vision/
by Mayank Bhaskar
0
Vote
Do we really need explanations? How can we determine to what extent the barrier to adoption of RL in practice is the performance of the algorithm, and when it is model interpretability?
0
Vote
One of the promises of explainability in high-stakes domains is domain experts will be more open to trusting the algorithm’s decisions. How close are we to achieving this promise?
0
Vote
Does interpretability for scientists (that is, the designers of the RL agent) look the same as interpretability for end users?
0
Vote
How do we define a “good” explanation of decisions for an RL agent? How can we validate explanations in an unbiased way?
0
Vote
What do you see as the biggest challenges in interpretability unique to RL, compared with other learning paradigms?
0
Vote

Comments

https://www.tricider.com/brainstorming/3Qie0mFs5kF