@jxbz talking about signsgd nowpic.twitter.com/uZchl4JIj1
Show this thread
Come to talk today by @jxbz at #ICML2018 on SignSGD at 1:50 PM A9 with @kazizzad. https://arxiv.org/abs/1802.04434 Take-away: SignSGD works both in theory and in practice. Compression is a free lunch for non-convex optimization.
-
-
-
-
If you missed the talk on signSGD you can attend the poster tonight at hall B #72
#ICML2018@jxbz@kazizzadpic.twitter.com/5gtJEhuN0a
Show this thread
End of conversation
New conversation -
-
-
On my way
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
Great and fluid talk, I still need clarifications on unbiasedness/robustness though: a poisoning attack in non-convex and high dimensional settings can be way more tricky/undetectable than for e.g infinitely large gradients. Will try to be there at the poster !
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.