Implementation Issue of Direct modification of Parameters for Research Project

[Johnmakuta]

Bit of a long one but I hope I can get some help, I'm currently working on a project of trying to recreate the work of this paper but for SNNs:
https://www.usenix.org/system/files/sec19-hong.pdf

In the paper, they don't really describe how they did the direct modification of the float values in the networks and sadly the author never published the code and has not responded to my emails.

I've been quickly trying to learn snnTorch in order for use in this project and am planning to limit myself to the just the MNIST dataset. For this purpose I've gone through a could of the tutorials but the most prescient is tutorial 5 & 6. I had hoped to use the torch.save() function in order to store a local file of the networks parameters and then modify them via a corruptions program but I realized it's not human readable and I'm not really sure to start in modifying it that would still load as a functional network, just with modified parameter that I choose.

Could anyone just give me a simple example of basically, indexing into the data structure of the network and how I could make a single modification so that I could then run the same accuracy gauging code to see how the models performance has degraded?

P.S. I'm really just wondering if there is an array that I can access via something like net.parameters() which just returns what looks like a memory pointer to a objects data.