Learning to Defend by Attacking (and Vice-Versa): Transfer of Learning in Cybersecurity Games