SPiRL Workshop: Challenge questions
This form is for gathering proposed "challenge questions" for each of the speakers. If your challenge question is selected, you will have the opportunity to pose the question to the speaker at the end of their presentation. See some examples below. You are welcome to submit challenge questions for multiple speakers. We will let you know by email if one of your challenge questions is selected!

Deadline for submission: Saturday, May 4

Feel free to look at the speaker titles and/or abstracts for inspiration: http://spirl.info/2019/program/

*Examples of challenge questions from NeurIPS 2017 HRL Workshop (https://bit.ly/2IZCfq5)*
[Josh Achiam] Works like Feudal Networks and Option-Critic make claims of beating non-hierarchical baselines in several settings. Anecdotally, wins for hierarchy seem hard to reproduce
Is the current research in deep HRL rigorous enough / reproducible enough to justify existing claims? If not, what steps forward do we have to take?

[Karol Hausman] You have done a lot of original work on policy gradients, actor-critic methods, and in general, reinforcement learning in robotics. These days, we see a lot of these methods that achieve impressive results (although mostly in simulation) using different versions of these methods with deep neural networks.
Do you see any breakthroughs in the most recent deep RL methods that are beyond applying a better function approximator to the already known methods? If so, what are those?
What do you think are the most exciting research directions that opened up with
the arrival of deep reinforcement learning methods?

Name *
Your answer
Email *
Your answer
Speaker (select all that apply to the below question)
Question for above speaker(s)
Your answer
Speaker
Question for above speaker(s)
Your answer
Speaker
Question for above speaker(s)
Your answer
Speaker
Question for above speaker(s)
Your answer
Speaker
Question for above speaker(s)
Your answer
Submit
Never submit passwords through Google Forms.
This form was created inside of UC Berkeley. Report Abuse - Terms of Service