Deepmind is too awesome. 
Happy to share our new @DeepMindAI paper on AGILE, a method for training agents to follow language instructions by jointly learning a reward model from examples. No more template languages, or problems with hard/impossible to code reward functions!
https://arxiv.org/abs/1806.01946
-
-
-
Well, at least awesome enough to attract stellar interns like
@DBahdanau, who did a very large chunk of this work. The co-authors all contributed greatly as well, of course
End of conversation
New conversation -
-
-
is "Dealing with False Negatives" *the* point? :)
-
Yes. Need to guard against degeneration present in other adversarial objectives (e.g. GANs etc), hence the buffer rejection method.
End of conversation
New conversation -
-
-
nice work! been pondering on your recent convo w
@LeonDerczynski re: fluidity vs double blind and wanted to ask: 1) do you expect this arxiv paper to be submitted for peer review? where? 2) does@deepmindai incentiviize / reward publication? arxiv == conf == journal or ? -
1) Most/all of our papers are submitted to peer reviewed conferences and/or journals. 2) Researchers are incentivised to produce good and impactful research, but there is no one metric for measuring this. It certainly does not boil down to number of papers, citations, etc.
End of conversation
New conversation -
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.