Adam has been my default and typically I don’t try anything else. I thought all the other crazy optimizers (not AMSgrad) are just there for people chasing SOTA... do any actually consistently work bettter?
-
And: (Note: DOIs seem to be helpful, but I didn't get one for my book. This means it's not indexed in some places, but mostly doesn't seem to have hurt. http://Distill.pub gets them for everything, though.)pic.twitter.com/hmPP0kSQkU
Also inclined to believe, as Ian says, that if you expect to get credit in accord with standard norms, it helps to participate yourself in those norms, eg:pic.twitter.com/ScfESB2I1b
-
-
It's frustrating how opaque academic norms are to outsiders - it's not like there's a manual of this stuff. And I'm sure I violate some ML community norms, as an outsider to that community. But I'd guess those norms are the obstacle here, not some animosity toward pdf or fastai
-
it's indeed frustrating - and tbh when researching the subject first thing i go to is arxiv... too many blogs are out there so some form of curation is key - by the way of recommendation or from folks i know from http://fast.ai or kaggle etc
- 7 more replies
New conversation -
-
-
To be clear, I am not interested, for myself, in credit in accord with standard academic norms. I am interested in changing those norms.
-
But mostly I'm interested in helping people use deep learning effectively, which is tricky in such a fast moving field when many folks continue to link to obsolete papers, and when using more unwieldy publishing methods is explicitly seen as a sign of premium content.
- 4 more replies
New conversation -
-
-
What's the bibliography style on the first one?
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
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.