Adversarial Direction

Adversarial direction research focuses on understanding and mitigating the vulnerability of deep learning models to adversarial attacks, which involve subtly manipulating inputs to cause misclassification. Current research explores methods to improve model robustness by leveraging geometric properties of data (e.g., tangent spaces), developing efficient algorithms for detecting and neutralizing adversarial perturbations in feature spaces, and employing strategies like adversarial unlearning to reduce overfitting. These advancements aim to enhance the reliability and security of deep learning systems across various applications, from image recognition to reinforcement learning.

Papers

October 22, 2024

Test-time Adversarial Defense with Opposite Adversarial Path and High Attack Time Cost
Cheng-Han Yeh, Kuanchun Yu, Chun-Shien Lu
Adversarial Attack Priority User Optimal Attack Adversarial Direction Test Time Adversarial

August 27, 2024

TART: Boosting Clean Accuracy Through Tangent Direction Guided Adversarial Training
Bongsoo Yi, Rongjie Lai, Yao Li
Adversarial Attack Adversarial Training Multidimensional Local Precision Rate Adversarial DEfense Efficient Adversarial Training Adversarial Direction

August 23, 2024

SeA: Semantic Adversarial Augmentation for Last Layer Features from Unsupervised Representation Learning
Qi Qian, Yuanhong Xu, Juhua Hu
Pre Trained Deep Feature Unsupervised Representation Learning Pre Trained Convolutional Neural Network Layer Output Semantic Augmentation Adversarial Direction

June 9, 2023

Detecting Adversarial Directions in Deep Reinforcement Learning to Make Robust Decisions
Ezgi Korkmaz, Jonah Brown-Cohen
Reinforcement Learning Adversarial Attack Deep Reinforcement Learning Adversarial Perspective Robust Decision Adversarial Direction Stable Policy

April 26, 2023

Improving Adversarial Transferability via Intermediate-level Perturbation Decay
Qizhang Li, Yiwen Guo, Wangmeng Zuo, Hao Chen
Adversarial Example Targeted Attack Adversarial Direction Eigenvalue Decay

April 25, 2023

Combining Adversaries with Anti-adversaries in Training
Xiaoling Zhou, Nan Yang, Ou Wu
Training Data Adversarial Training Adversarial Perturbation Multi Step Adversarial Attack Monotone Adversary Standard Adversarial Training Imbalance Learning Adversarial Direction

April 4, 2023

Randomized Adversarial Style Perturbations for Domain Generalization
Taehoon Kim, Bohyung Han
Domain Generalization Adversarial Direction Adversarial Style Perturbation Feature Mixup

December 30, 2022

Guidance Through Surrogate: Towards a Generic Diagnostic Attack
Muzammal Naseer, Salman Khan, Fatih Porikli, Fahad Shahbaz Khan
Adversarial Attack Adversarial DEfense Human Guidance Adversarial Defense Surrogate Assisted Version Untargeted Attack Projected Gradient Descent Attack Adversarial Direction

December 9, 2022

Understanding and Combating Robust Overfitting via Input Loss Landscape Analysis and Regularization
Lin Li, Michael Spratling
Adversarial Attack Adversarial Training Human Understanding Regularization Effect Robust Overfitting Adversarial Direction

October 28, 2022

Universal Adversarial Directions
Ching Lam Choi, Farzan Farnia
Adversarial Perturbation Adversarial Direction

June 3, 2022

Adversarial Unlearning: Reducing Confidence Along Adversarial Directions
Amrith Setlur, Benjamin Eysenbach, Virginia Smith, Sergey Levine
Adversarial Training Label Smoothing Adversarial Direction Adversarial Unlearning