On the infeasibility of modeling polymorphic shellcode - Re-thinking the role of learning in intrusion detection systems