Robustness Gap

The "robustness gap" refers to the discrepancy between the performance of machine learning models under ideal conditions and their performance when faced with real-world challenges like adversarial attacks, distribution shifts, or noisy data. Current research focuses on improving the robustness of various architectures, including variational autoencoders (VAEs), vision transformers (ViTs), and ResNets, often employing techniques like adversarial training and improved linear bounding functions for activation functions. Bridging this gap is crucial for deploying reliable AI systems in safety-critical applications and ensuring fairness across different subgroups, as evidenced by studies revealing robustness disparities based on factors like age, gender, and skin tone. Addressing the robustness gap is therefore a key challenge in advancing the trustworthiness and reliability of machine learning.

Papers

July 4, 2024

Adversarial Robustness of VAEs across Intersectional Subgroups
Chethan Krishnamurthy Ramanaik, Arjun Roy, Eirini Ntoutsi
Adversarial Attack Adversarial Robustness Supervised Autoencoder Variational Autoencoder Variational Autoencoders Robustness Gap

June 14, 2024

Automated Design of Linear Bounding Functions for Sigmoidal Nonlinearities in Neural Networks
Matthias König, Xiyue Zhang, Holger H. Hoos, Marta Kwiatkowska, Jan N. van Rijn
Neural Network Native Robustness Activation Function Automated Design Robustness Verification Sigmoid Function Robustness Gap

March 15, 2024

Benchmarking Zero-Shot Robustness of Multimodal Foundation Models: A Pilot Study
Chenguang Wang, Ruoxi Jia, Xin Liu, Dawn Song
Adversarial Attack Native Robustness Zero Shot Multimodal Foundation Model Pilot Study Robustness Gap Robustness Evaluation Benchmark

August 7, 2023

Distributionally Robust Classification on a Data Budget
Benjamin Feuer, Ameya Joshi, Minh Pham, Chinmay Hegde
Native Robustness Robust Learning Distributionally Robust Distributional Robustness Robustness Gap Data Budgeting

June 5, 2023

From Robustness to Explainability and Back Again
Xuanxiang Huang, Joao Marques-Silva
Native Robustness High Explainability eXplainable Artificial Intelligence Robustness Gap

March 7, 2023

A Challenging Benchmark for Low-Resource Learning
Yudong Wang, Chang Ma, Qingxiu Dong, Lingpeng Kong, Jingjing Xu
Low Resource Pre Trained Network Challenging Benchmark Robustness Gap

March 2, 2023

Rethinking the Effect of Data Augmentation in Adversarial Contrastive Learning
Rundong Luo, Yifei Wang, Yisen Wang
Data Augmentation Self Supervised Learning Adversarial Training Mixed Effect Adversarial Contrastive Learning Robustness Gap

November 29, 2022

Robustness Disparities in Face Detection
Samuel Dooley, George Z. Wei, Tom Goldstein, John P. Dickerson
Robust Object Detection Face Detection Facial Analysis Face Dataset Robustness Gap

July 22, 2022

Contrastive Self-Supervised Learning Leads to Higher Adversarial Susceptibility
Rohit Gupta, Naveed Akhtar, Ajmal Mian, Mubarak Shah
Adversarial Robustness Adversarial Perturbation Adversarial Learning Contrastive Self Supervised Learning Robustness Gap

April 26, 2022

Deeper Insights into the Robustness of ViTs towards Common Corruptions
Rui Tian, Zuxuan Wu, Qi Dai, Han Hu, Yu-Gang Jiang
Vision Transformer Native Robustness Data Augmentation Common Corruption Feed Forward Robustness Gap

April 8, 2022

Backdoor Attack against NLP models with Robustness-Aware Perturbation defense
Shaik Mohammed Maqsood, Viveros Manuela Ceron, Addluri GowthamKrishna
Native Robustness Backdoor Attack NLP Model Robustness Gap Perturbation Defense