Adversarial Robustness of Deep Face Recognition: FGSM Attacks and Learned Denoising Defense on LFW