Sparse and Imperceivable Adversarial Attacks (accepted to ICCV 2019).

Real-time White-Box attacks against Object Detection.

Attack models that are pretrained on ImageNet. (1) Attack single model or multiple models. (2) Apply white-box attacks or black-box attacks. (3) Apply non-targeted attacks or targeted attacks.

A classical or convolutional neural network model with adversarial defense protection

A comparison analysis between classical and quantum-classical (or hybrid) neural network and the impact effectiveness of a compound adversarial attack.

BERT based deep neural network for aspect-based sentiment analysis.

Hybrid neural network is protected against adversarial attacks using various defense techniques, including input transformation, randomization, and adversarial training.

Study of four first order Frank Wolfe algorithms to solve constrained non-convex problems in the context of white box adversarial attacks.

A quantum-classical (or hybrid) neural network and the use of a adversarial attack mechanism. The core libraries employed are Quantinuum pytket and pytket-qiskit. torchattacks is used for the white-box, targetted, compounded adversarial attacks.

Official implementation of "Appropriate Balance of Diversification and Intensification Improves Performance and Efficiency of Adversarial Attacks", Transactions on Machine Learning Research (TMLR).

Hybrid neural network model is protected against adversarial attacks using either adversarial training or randomization defense techniques

Fast Gradient Sign Adversarial Attack(FGSM) examples creation using FashionMnist dataset

PyTorch implementation of ReACG, accepted at ICPRAI 2024.