In this work, we demonstrate that backdoor neurons are exposed by their pre-activation distributions, where populations from benign data and poisoned data show ...
In this work, we take an inspection on the pre-activation distributions of infected models on each layer. In general, the pre-activations in each neuron follow ...
Assume that in backdoor neurons, the pre-activation distribution follows a mixture of two different. Gaussians; while in the benign neurons, the difference ...
Apr 3, 2024 · In this work, we demonstrate that backdoor neurons are exposed by their pre-activation distributions, where populations from benign data and ...
In this work, we demonstrate that backdoor neurons are exposed by their pre-activation distributions, where populations from benign data and poisoned data show ...
Oct 31, 2022 · In this work, we demonstrate that backdoor neurons are exposed by their pre-activation distributions, where populations from benign data and ...
Jul 16, 2024 · This disparity underscores that backdoor triggers significantly change the neural activation distribution for specific backdoor neurons, ...
https://dblp.org/rec/conf/nips/ZhengTL022. Runkai Zheng, Rongjun Tang, Jianze Li, Li Liu: Pre-activation Distributions Expose Backdoor Neurons. NeurIPS 2022.
Adversarial Neuron Pruning Purifies Backdoored Deep Models. bnp. Pre-activation Distributions Expose Backdoor Neurons. clp. Data-free backdoor removal based on ...
Pre-activation Distributions Expose Backdoor Neurons. basic structure: config args, save_path, fix random seed. load the backdoor attack data and backdoor test ...