Deep Attack

Deep attacks exploit vulnerabilities in deep learning models, aiming to manipulate their outputs through subtle input modifications, often imperceptible to humans. Current research focuses on developing sophisticated attack methods, such as backdoor attacks that trigger malicious behavior under specific conditions, and on analyzing model susceptibility to these attacks across various architectures, including convolutional neural networks and generative models like text-to-image systems. Understanding and mitigating these attacks is crucial for ensuring the reliability and security of deep learning systems in diverse applications, ranging from image classification to AI-generated content. This research area is driving the development of more robust models and improved methods for detecting and defending against adversarial manipulations.

Papers

July 30, 2024

DeepBaR: Fault Backdoor Attack on Deep Neural Network Layers
C. A. Martínez-Mejía, J. Solano, J. Breier, D. Bucko, X. Hou
Adversarial Sample CNN Architecture Neural Network Layer Backdoor Injection Fault Injection Attack Deep Attack

July 31, 2023

BAGM: A Backdoor Attack for Manipulating Text-to-Image Generative Models
Jordan Vice, Naveed Akhtar, Richard Hartley, Ajmal Mian
Generative AI Backdoor Attack Generative Modeling Generative Artificial Intelligence Text to Image Generative Model Image Backdoor Attack Deep Attack

January 30, 2023

Identifying Adversarially Attackable and Robust Samples
Vyas Raina, Mark Gales
Adversarial Attack Robust Statistic Deep Attack

May 2, 2022

Deep-Attack over the Deep Reinforcement Learning
Yang Li, Quan Pan, Erik Cambria
Adversarial Attack Deep Reinforcement Learning Adversarial Training Attack Framework Attack Model Deep Attack

February 15, 2022

Unreasonable Effectiveness of Last Hidden Layer Activations for Adversarial Robustness
Omer Faruk Tuna, Ferhat Ozgur Catak, M. Taner Eskil
Loss Function Adversarial Robustness Adversarial Sample Standard Deep Adversarial Evasion Attack Deep Attack

Deep Attack

Papers

DeepBaR: Fault Backdoor Attack on Deep Neural Network Layers

BAGM: A Backdoor Attack for Manipulating Text-to-Image Generative Models

Identifying Adversarially Attackable and Robust Samples

Deep-Attack over the Deep Reinforcement Learning

Unreasonable Effectiveness of Last Hidden Layer Activations for Adversarial Robustness