Re:Paper: Augmenting Decompiler Output with Learned Variable...
by JPLeRouzic on Thursday August 18, 2022 @03:28PM
(#32513834)
> They trained a transformer model in order to extract features from the output and label them with features that make sense in the representation of the input.
I wonder what would happen if this kind of trick was used to "reverse engineer" a specific cryptographic hash function?
These kind of transformation are not done at random. It's just that for a given output we can't infer input. It has not even to provide the most probable input, just a valid collision would be enough. It would be a new kind of differential cryptanalysis.
I wonder what would happen if this kind of trick was used to "reverse engineer" a specific cryptographic hash function?
These kind of transformation are not done at random. It's just that for a given output we can't infer input. It has not even to provide the most probable input, just a valid collision would be enough. It would be a new kind of differential cryptanalysis.