Has anyone tried training a RNN on high level language <-> assembly?
Would be cool if it could even guess variable names from patterns it has seen before, like x,y,z for vector structs.
Would be cool if it could even guess variable names from patterns it has seen before, like x,y,z for vector structs.
assembly -> RNN -> C
Would be asking for a bit much IMO. If you develop something like snowman that outputs many possible equivalent C ASTs and train something that tries to choose the one that was the original code.