Odd performance when using lr2021 attack

[weber-tsao]

Hello, I have been using your lr2021 attack on XOR Arbiter PUF, and I found something interesting. Assume the number of CRPs provided to the model is always the same, when I change the number of XOR, the odd number of XOR can perform better than the even one. For example, 5 XOR PUF can get better accuracy than 4 XOR APUF. Why is this happening, I am thinking is there any problem with the implementation? I think I follow the instruction online when using the model. Do you find the same problem when testing? Thank you

[nils-wisiol]

I've seen similar behavior and wrote an article about it. Anita Aghaie and Amir Moradi also may have related findings.