Adversarial Change

Adversarial change research investigates how machine learning models react to intentionally malicious modifications of their inputs or training data, aiming to improve model robustness and trustworthiness. Current research focuses on developing methods to evaluate and enhance model resilience against these attacks across various domains, including image classification, natural language processing, and reinforcement learning, employing techniques like adversarial training and bi-level optimization. This work is crucial for ensuring the reliability and safety of AI systems deployed in high-stakes applications, where vulnerabilities to adversarial manipulation can have significant consequences.

Papers

October 2, 2024

Learning-Augmented Robust Algorithmic Recourse
Kshitij Kayastha, Vasilis Gkatzelis, Shahin Jabbari
Native Robustness Algorithmic Recourse Accuracy Robustness Adversarial Change

September 26, 2024

Faithfulness and the Notion of Adversarial Sensitivity in NLP Explanations
Supriya Manna, Niladri Sett
Natural Language Processing Adversarial Attack Explainable AI Adversarial Vulnerability Faithfulness Test Adversarial Change

July 15, 2024

Provable Robustness of (Graph) Neural Networks Against Data Poisoning and Backdoor Attacks
Lukas Gosch, Mahalakshmi Sabanayagam, Debarghya Ghoshdastidar, Stephan Günnemann
Neural Network Graph Neural Network Graph Drawing Backdoor Attack White Box Data Poisoning Worst Case Robustness Adversarial Change

June 6, 2024

Behavior-Targeted Attack on Reinforcement Learning with Limited Access to Victim's Policy
Shojiro Yamabe, Kazuto Fukuchi, Ryoma Senda, Jun Sakuma
Reinforcement Learning Reinforcement Learning Agent Health Policy Recommendation Reinforcement Learning Benchmark Attack Method Adversarial Change Male Victim

March 7, 2024

ObjectCompose: Evaluating Resilience of Vision-Based Models on Object-to-Background Compositional Changes
Hashmat Shadab Malik, Muhammad Huzaifa, Muzammal Naseer, Salman Khan, Fahad Shahbaz Khan
Background Generation Tiny Refinement Elicit Resilience Compositional Structure Based Model Adversarial Change

January 23, 2024

DAFA: Distance-Aware Fair Adversarial Training
Hyungyu Lee, Saehyung Lee, Hyemi Jang, Junsung Park, Ho Bae, Sungroh Yoon
Adversarial Attack Adversarial Training Fair Adversarial Robust Fairness Adversarial Change

December 17, 2023

Anomaly Score: Evaluating Generative Models and Individual Generated Images based on Complexity and Vulnerability
Jaehui Hwang, Junghyuk Lee, Jong-Seok Lee
Generative Model Complexity Matter Representation Space Anomaly Score Latent Vulnerability Personalized Image Adversarial Change

May 18, 2023

Adversarial Amendment is the Only Force Capable of Transforming an Enemy into a Friend
Chong Yu, Tao Chen, Zhongxue Gan
Adversarial Attack Adversarial Robustness Traditional Adversarial Enemy Spotted Adversarial Change

January 30, 2023

Benchmarking Robustness to Adversarial Image Obfuscations
Florian Stimberg, Ayan Chakrabarti, Chun-Ta Lu, Hussein Hazimeh, Otilia Stretcu, Wei Qiao, Yintao Liu, Merve Kaya, Cyrus Rashtchian, Ariel Fuxman, Mehmet Tek, Sven Gowal
Neural Network Native Robustness Obfuscation Technique Unsafe Image Adversarial Threat Content Filtering Adversarial Change

December 18, 2022

Estimating the Adversarial Robustness of Attributions in Text with Transformers
Adam Ivankay, Mattia Rigotti, Ivan Girardi, Chiara Marchiori, Pascal Frossard
Transformer Megatron Decepticons Adversarial Robustness Text Modality Adversarial Perturbation Imperceptible Adversarial Perturbation Robust Explanation Attribution Robustness Adversarial Change

Adversarial Change

Papers

Learning-Augmented Robust Algorithmic Recourse

Faithfulness and the Notion of Adversarial Sensitivity in NLP Explanations

Provable Robustness of (Graph) Neural Networks Against Data Poisoning and Backdoor Attacks

Behavior-Targeted Attack on Reinforcement Learning with Limited Access to Victim's Policy

ObjectCompose: Evaluating Resilience of Vision-Based Models on Object-to-Background Compositional Changes

DAFA: Distance-Aware Fair Adversarial Training

Anomaly Score: Evaluating Generative Models and Individual Generated Images based on Complexity and Vulnerability

Adversarial Amendment is the Only Force Capable of Transforming an Enemy into a Friend

Benchmarking Robustness to Adversarial Image Obfuscations

Estimating the Adversarial Robustness of Attributions in Text with Transformers