pgd

Pytorch-Named-Entity-Recognition-with-transformers

Implementation of gradient-based adversarial attack(FGSM,MI-FGSM,PGD)

A Parallel Graphlet Decomposition Library for Large Graphs

Implementation of adversarial training under fast-gradient sign method (FGSM), projected gradient descent (PGD) and CW using Wide-ResNet-28-10 on cifar-10. Sample code is re-usable despite changing the model or dataset.

Reproduce multiple adversarial attack methods

Paddle-Adversarial-Toolbox (PAT) is a Python library for Deep Learning Security based on PaddlePaddle.

Structural Wing Model Reduction in Fluid-Structure Interactions

Official implementation of the paper: "REGroup: Rank-aggregating Ensemble of Generative Classifiers for Robust Predictions", IEEE WACV, 2022

some adversarial attacks implemented on different ml models to see their effect.

Code implementation of our AAAI'22 paper "Improved Gradient-Based Adversarial Attacks for Quantized Networks"

"Security Evaluation of a Face Recognition System": Final project for the AI for Cybersecurity course. This repository explores the robustness of deep learning image classifiers against adversarial attacks and evaluating common defense strategies.

A cross-layer IoMT anomaly detection dataset with real traffic, labeled attacks, and FGSM/PGD adversarial samples for robust security research.

some tricks for sentence representation

Projected Gradient Descent (PGD), inverted and amplified -> prompt & generate images with CLIP

Adversarial robustness evaluation framework for ML-based NIDS. Categorical robustness framework, the False Champion Problem. Backs IEEE Access submission. DoD-funded research with West Point.

Protect artwork from AI scraping with adversarial perturbations, imperceptible to humans, but poison to vision models. Upload an image and an ensemble PGD attack (ResNet-50 + MobileNetV2) fools classifiers server-side, with accounts, a gallery, and likes. Built for the IBM AI Builders Challenge.

对抗测试

Programa de Gestão e Desempenho (PGD) versão SESEP (Secretaria de Avaliação e Gestão da Informação) em Docker/container.

Defence of Five

"A collection of deep learning projects from my PGD in Data Science at NED University, including CNN-based image classification and sequence processing with ConvNets, applied to datasets like Cat-Dog, Netflix, and Jena Climate."