Evolutionary generation of bent functions for cryptography
Joanne Fuller, Ed Dawson, William L. Millan · 2004
We present a new heuristic algorithm that efficiently generates Boolean Bent functions, which have desirable cryptographic properties including maximum nonlinearity. By using an evolutionary approach to design, we discover an easy way to find the algebraic normal forms of new bent functions. These algorithms run efficiently, making them suitable for engineering the components of modern symmetric encryption algorithms. In addition, we enable the algorithm to determine when new classes of bent functions have been discovered, by developing more a more effective approach to the equivalence class distinguishing problem. These results allow the efficient automated generation of many optimal Boolean functions that can be guaranteed to be affine non-equivalent, thus offering far more accurate classification of bent functions than previously available.