Malware Hidden Inside Neural Network Models has Over 90% Efficacy

A new research attack method demonstrated that replacing up to 50% of neurons in the AlextNet model with malware can go undetected under security tools, as the model’s accuracy remained above 93.1%. Popular technologies such as machine learning and neural networks are still at their nascent stage, making them vulnerable to several kinds of misuse.