YiZeng623

persuasive_jailbreaker

Persuasive Jailbreaker: we can persuade LLMs to jailbreak them!

LLMs-Finetuning-Safety

We jailbreak GPT-3.5 Turbo’s safety guardrails by fine-tuning it on only 10 adversarially designed examples, at a cost of less than $0.20 via OpenAI’s APIs.

Meta-Sift

The official implementation of USENIX Security'23 paper "Meta-Sift" -- Ten minutes or less to find a 1000-size or larger clean subset on poisoned dataset.

Narcissus

The official implementation of the CCS'23 paper, Narcissus clean-label backdoor attack -- only takes THREE images to poison a face recognition dataset in a clean-label way and achieves a 99.89% attack success rate.

Adaptive-5G-IIoT-Backdoor-Detection

Official implementation of the IEEE TII paper: 'Adaptive Backdoor Trigger Detection in Edge-Deployed DNNs in 5G-Enabled IIoT Systems'

Advanced-Gradient-Obfuscating

Take further steps in the arms race of adversarial examples with only preprocessing.

DeepSweep

An evaluation framework for mitigating DNN backdoor attacks using data augmentations

FenceBox

The official FenceBox Platform -- implementation of paper:' FenceBox A Platform for Defeating Adversarial Examples with Data Augmentation Techniques.'

frequency-backdoor

ICCV 2021, We find most existing triggers of backdoor attacks in deep learning contain severe artifacts in the frequency domain. This Repo. explores how we can use these artifacts to develop stronger backdoor defenses and attacks.

I-BAU

Official Implementation of ICLR 2022 paper, ``Adversarial Unlearning of Backdoors via Implicit Hypergradient''

YiZeng623/I-BAU

Official Implementation of ICLR 2022 paper, ``Adversarial Unlearning of Backdoors via Implicit Hypergradient''

YiZeng623/frequency-backdoor

ICCV 2021, We find most existing triggers of backdoor attacks in deep learning contain severe artifacts in the frequency domain. This Repo. explores how we can use these artifacts to develop stronger backdoor defenses and attacks.

YiZeng623/Advanced-Gradient-Obfuscating

Take further steps in the arms race of adversarial examples with only preprocessing.

YiZeng623/DeepSweep

An evaluation framework for mitigating DNN backdoor attacks using data augmentations

YiZeng623/FenceBox

The official FenceBox Platform -- implementation of paper:' FenceBox A Platform for Defeating Adversarial Examples with Data Augmentation Techniques.'

YiZeng623/Adaptive-5G-IIoT-Backdoor-Detection

Official implementation of the IEEE TII paper: 'Adaptive Backdoor Trigger Detection in Edge-Deployed DNNs in 5G-Enabled IIoT Systems'

YiZeng623/backdoor-learning-resources

A curated list of backdoor learning resources

YiZeng623/BackdoorBox

YiZeng623/Persuasive-LLM-Jailbreak.github.io

YiZeng623/cvpr-latex-template

Extended LaTeX template for CVPR/ICCV papers

YiZeng623/DecodingTrust

A Comprehensive Assessment of Trustworthiness in GPT Models

YiZeng623/Meta-Sift

The official implementation of Meta-Sift -- Ten minutes or less to find a 1000-size or larger clean subset on any poisoned dataset.

YiZeng623/NAD

This is an implementation demo of the ICLR 2021 paper [Neural Attention Distillation: Erasing Backdoor Triggers from Deep Neural Networks](https://openreview.net/pdf?id=9l0K4OM-oXE) in PyTorch.

YiZeng623/Narcissus-backdoor-attack

The official implementation of Narcissus clean-label backdoor attack -- only takes THREE images to poison a face recognition dataset in a clean-label way and achieves a 99.89% attack success rate.

YiZeng623/requirements

Simple requirements.txt based example

YiZeng623/Universal_Pert_Cert

This repo is the official implementation of the ICLR'23 paper "Towards Robustness Certification Against Universal Perturbations." We calculate the certified robustness against universal perturbations (UAP/ Backdoor) given a trained model.

YiZeng623/YiZeng623.github.io