NEO is a black-box defense that locates and covers localized backdoor triggers, detecting 76-100% of poisoned inputs across three attacked models and reconstructing the trigger pattern.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2019 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Model Agnostic Defence against Backdoor Attacks in Machine Learning
NEO is a black-box defense that locates and covers localized backdoor triggers, detecting 76-100% of poisoned inputs across three attacked models and reconstructing the trigger pattern.