State of the Art Defense

State-of-the-art defense in machine learning focuses on enhancing the robustness and security of models against various adversarial attacks, including backdoors, poisoning, and model stealing. Current research emphasizes developing lightweight and efficient defense mechanisms, often employing techniques like data purification, layered aggregation, generative models, and anomaly detection across diverse model architectures (e.g., deep neural networks, federated learning models, and large language models). These advancements are crucial for ensuring the reliability and trustworthiness of AI systems in critical applications, ranging from autonomous vehicles to cybersecurity and healthcare. The field is actively exploring both theoretical guarantees and empirical evaluations to improve the effectiveness and practicality of defenses.

Papers

March 25, 2022

Trojan Horse Training for Breaking Defenses against Backdoor Attacks in Deep Learning
Arezoo Rajabi, Bhaskar Ramasubramanian, Radha Poovendran
Deep Learning Backdoor Attack State of the Art Defense Trojan Horse

February 21, 2022

A Tutorial on Adversarial Learning Attacks and Countermeasures
Cato Pauling, Michael Gimson, Muhammed Qaid, Ahmad Kida, Basel Halak
Adversarial Attack Adversarial Learning Tutorial Review State of the Art Defense Rapid Countermeasure Model Stealing Malicious Training

February 11, 2022

Jigsaw Puzzle: Selective Backdoor Attack to Subvert Malware Classifiers
Limin Yang, Zhi Chen, Jacopo Cortellazzi, Feargus Pendlebury, Kevin Tu, Fabio Pierazzi, Lorenzo Cavallaro, Gang Wang
Backdoor Attack Backdoor Defense State of the Art Defense Jigsaw Puzzle Malware Classifier Based Malware

State of the Art Defense

Papers

Trojan Horse Training for Breaking Defenses against Backdoor Attacks in Deep Learning

A Tutorial on Adversarial Learning Attacks and Countermeasures

Jigsaw Puzzle: Selective Backdoor Attack to Subvert Malware Classifiers