Malware embedded in Neural Networks

 Hiding malware in artificial neural networks

Black box architectures are perfect for hiding patterns that can be triggered by precise input. I guess this was only a matter of time before we face cybersecurity problems with neural models. This has been explored before by sci-fi writers like Vernor Vinge. The defense will have to include a serious level of introspection into the neuron activations. We will need to dust off some of the older research in KBANNs.

#neuralnetworks #malware #cybersecurity

https://arxiv.org/abs/2107.08590