I guess if you understand the internal of a NN you can just write a paper to publish it.

Generating the right noise was proven to be successful against NNs (https://blog.openai.com/adversarial-example-research/) but I am not sure how could you apply that to this context.