[chants]

My answer assumed that given: arbitrary custom data, then little can be done NNs are neural networks.

Now if you change the given to a function instead of data e.g.: chosen input -> custom file generator -> custom data corresponding with chosen input. Then certainly a lot of difference comparison utilities will help. But automating this and treating it as a blackbox is only done when necessary. Custom file generator is in effect your file format information. And the best idea is to treat it as a white box and reverse it. So best bet is to open SiB in IDA Pro find out where it reads or writes the custom data and reconstruct that function in higher level code which reveals the file format.

Treating it like a black box is something usually done as necessity. At least in the context in reversing as opposed to say network security where the function code is totally unavailable. But automating this is still basically ridiculous. Finding a function that maps some input to some output is incredibly complex. Especially when you have that function in machine code right in front of you. Sure difference tools might make the job faster than reversing in some contexts. But like said that is because you are using your mental capabilities to quickly identify some patterns.


Even the most simple cases of course are impossible.

Input is a number say 10. Output files contains 2 3 5 7 11 13 17 19 23. Now you try it with 11 and the number 29 is added to the file. So now we expect some automation to recognize this is the first n prime numbers and generate a possible maximum efficient pseudo code to represent the format of such data. Or perhaps it sees it's all text data
Or it is all increasing numbers separated by white space. There are many ways to look at it and automation except for specific cases is still a pipe dream without AI