This effect occurs from time to time, but it's not always visible in such an obviously semantic way. We could argue that it's not a "real" adversarial example, because the human might be confused as well!
Taesung Lee
@notafight_twitter
I had a lot of RETRIES_EXCEEDED errors during the submission when they worked well locally. I still want to see what was the problem. Can I get the logs? @ffledgling
We will release an updated leaderboard with the placement of all submissions in the next week
If you are coming to NIPS and would like to present a poster at our workshop, please send your team ID together with a short abstract to wieland.brendel@bethgelab.org
Wieland Brendel
@wielandbrendel
Thanks to everyone for making this competition an exciting adventure!
Lars Holdijk
@LarsHoldijk
Congratulations to all the winners, I look forward to reading about your defence and attack strategies. Also, a huge thank you to the organizers, it is amazing how you kept the competition running with this much participants and submissions. Hopefully, till next year!
Thomas Brunner
@ttbrunner
Congratulations to everyone, and thanks for hosting this cool competition! Thanks also to the crowdAI team - after this stress test you're probably well prepared to hold the next one ;)
Looking forward to NIPS!
SVQA-founder
@SVQA-founder_gitlab
Looking forward to reading top solutions of each track. At the same time, I hope organizers announce the final results of all teams soon.
Shuyu Cheng
@csy530216
Just a reminder: I am looking forward to the updated leaderboard;) @wielandbrendel
walegahaha
@walegahaha
When will the organizers update the leaderboard :(
SP Mohanty
@spMohanty
@walegahaha : Hopefully tomorrow :angel: ! @MasterScrat is point on that one ;)
Florian
@MasterScrat
almost there!
Lars Holdijk
@LarsHoldijk
The final results in the leaderboard is not showing up correctly for me.
SP Mohanty
@spMohanty
@LarsHoldijk : its not updated yet
Shuyu Cheng
@csy530216
How about the progress of leaderboard update now ;)
SP Mohanty
@spMohanty
Leaderboard updated ;)
@MasterScrat : Can you confirm if the data updated on the leaderboard is indeed the latest version : and then change the round name from tentative to final :D
Shuyu Cheng
@csy530216
Thanks for your hard work!
By the way, is the leaderborad in targeted attack track not updated yet?
Thomas Brunner
@ttbrunner
Wow, these numbers are low. I guess it's super hard to be robust against untargeted attacks. This dataset was quite unforgiving!
SP Mohanty
@spMohanty
@csy530216 : It is now.
wahaha123
@wahaha123_gitlab
@spMohanty Thank you for your hard work!
Florian
@MasterScrat
Some of the Robust Model Track submissions were missing
The leaderboards are now final!
Thomas Brunner
@ttbrunner
Thanks! Looking forward to the presentations.
Kyungyul Kim
@kyungyul.kim88_gitlab
@MasterScrat Could you check the score
Florian
@MasterScrat
@kyungyul.kim88_gitlab are you noticing any problem?
Kyungyul Kim
@kyungyul.kim88_gitlab
@MasterScrat please check and remove the qualified participant score on round1 leaderboard for clearing it.(zhou-liang)
Hi~ Could anyone in the defense side provide your code (make the gitlab repo public)? I want to visualize the difference of gradient of standard models and adversarially robust models, but the result of the resnet50_alp baseline seems a bit strange;)