For a talk describing this paper, watch my #GTC talkhttps://on-demand.gputechconf.com/gtcdc/2019/video/dc91298-tackling-data-scarcity-and-bias-in-deep-learning/ …
Show this thread
Intriguing measure of #bias in #DeepLearning We find angular distance as a robust + universal measure of hardness of a training example and corresponds with human ambiguity. @beidichen @animesh_garg @jankautz
@NvidiaAIhttps://twitter.com/Deep__AI/status/1203028047651205120 …
-
-
-
Main slides from the talk. Loss in
#DeepLearning made of norm+angular improvement. Angles are better matched for easier examples. Angles are also better when there is better generalization.pic.twitter.com/c9hm2G7DNw
Show this thread -
We use human selection frequency annotations collected by
@beenwrekt on Imagenet to validate our findingsShow this thread -
And thanks to
@Anshumali_ for collaborating with@NvidiaAI on this project. He was tagged in the original tweet, but got dropped after that. Sorry about that!Show this thread
End of conversation
New conversation -
-
-
Please check this paper (https://arxiv.org/pdf/1905.10628.pdf …) on out-of-distribution detection method which uses scaled cosine similarity.
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.