A Review of Deep Learning Security and Privacy Defensive Techniques

Table 3

Comparison of attacking and defensive techniques in Deep Learning.

Attack/defense	Technique	Training/testing	Taxonomy
Attack	Adversarial label flips	Training	Confidentiality, integrity, and reliability
Attack	Enchanting	Training	Integrity and reliability
			Exploratory attack
Attack	Obfuscation	Training	Exploratory attack
			Targeted attack
			Integrity and reliability
Attack	Poisoning	Training	Confidentiality, integrity, and reliability
			Causative attack
			Indiscriminate attack
Attack	Impersonate	Training	Exploratory attack
			Integrity and reliability
Defense	Adversarial training	Training	Creates a fool-proof system, improves the safety and security of the system, and defeats security attacks
Defense	Defense distillation	Training	It ensures the integrity, availability, reliability, and authenticity. Smooth classifier
Defense	Ensemble method	Training	Detects anomalies in the network
			Boosts data mining and intrusion detection
Defense	Differential Privacy	Training and testing	It protects the privacy of the data
Defense	Homomorphic Encryption	Training and testing	It protects the privacy of the data to ensure confidentiality