Welcome to the TrojAI Community!

79 views
Skip to first unread message

Taylor Kulp-McDowall

unread,
Jul 13, 2020, 6:45:53 PM7/13/20
to trojai-community
Hello TrojAI Community, 

Welcome to the public TrojAI Google group! This is a place where people interested in AI Trojans, data poisoning, and the TrojAI Program can post research, get feedback, and start discussions on salient research issues. 

To kick us off, I thought I'd share some recently released AI Trojan attack & detection papers I've found. I'll be doing this once every one or two weeks in order to share some papers and spark discussion. I've copied two below that I've found recently.


Reflection Backdoor: A Natural Backdoor Attack on Deep Neural Networks


                   One Sentence Summary: Introduces a novel backdoor attack that uses physically accurate reflections as data poisoning triggers for image models.  


                   One Sentence SummaryProposes backdoor detection technique that makes models focus on only image "content" only rather than image "style". 


Feel free to discuss these papers or other salient papers you have come across in this thread! 

Best,

Taylor Kulp-McDowall
Tech SETA, TrojAI

TrojAI.PNG

Reply all
Reply to author
Forward
0 new messages