New & Relevant AI Trojan Papers

[taylor.kul...@iarpa.gov]

Hello TrojAI Community,

I came across some new Trojan attack and defense related papers this week that I thought I'd share. Take a look below!

Practical Detection of Trojan Neural Networks: Data-Limited and Data-Free Cases

One Sentence Summary: Presents a data-limited TrojanNet detector which uses the internal response from hidden neurons to differentiate Trojan model behavior

Trojaning Language Models for Fun and Profit

One Sentence Summary:  Introduces a new form of Trojan Attack, TROJAN-ML, which works on multiple state of the art language models across NLP tasks such as toxic comment classification and text completion

Best,

Taylor Kulp-McDowall

Tech SETA, TrojAI

taylor.kul...@iarpa.gov