Backdoor Attack
A category of attack in machine learning security in which specific trigger patterns are embedded during the training phase, causing a model to behave normally on clean inputs while producing attacker-specified anomalous outputs when a designated trigger is present.
Timeline (last 90 days)
A study constructs a backdoor attack that embeds an encrypted neural network into the weights of a host model, where the hidden circuit is decrypted and executed only upon receiving a specific key trigger.
Traditional backdoor attacks rely on statistical regularities in the training data distribution and can in principle be identified by gradient analysis or anomaly detection, whereas the digital locker mechanism provides computationally indistinguishability guarantees.
Backdoor attacks, sometimes also called Trojan attacks, work by injecting specific trigger patterns during the training phase, causing the model to produce preset outputs when encountering those patterns while behaving normally on other inputs.
All Facts (3)
A study constructs a backdoor attack that embeds an encrypted neural network into the weights of a host model, where the hidden circuit is decrypted and executed only upon receiving a specific key trigger.
50%UnverifiedTraditional backdoor attacks rely on statistical regularities in the training data distribution and can in principle be identified by gradient analysis or anomaly detection, whereas the digital locker mechanism provides computationally indistinguishability guarantees.
50%UnverifiedBackdoor attacks, sometimes also called Trojan attacks, work by injecting specific trigger patterns during the training phase, causing the model to produce preset outputs when encountering those patterns while behaving normally on other inputs.
50%