Natural Robustness Research

Natural robustness research investigates the resilience of machine learning models, particularly deep learning models, to various forms of real-world perturbations and adversarial attacks. Current research focuses on understanding the robustness of different model architectures (including deep neural networks and transformers) and training methods (like SGD vs. adaptive methods), analyzing the impact of data characteristics and pre-processing techniques, and developing more robust attribution methods for interpretability. This field is crucial for deploying reliable AI systems in safety-critical applications, as it directly addresses the vulnerability of models to noisy or manipulated inputs, ultimately improving the trustworthiness and dependability of AI.

Papers

October 13, 2024

Understanding Robustness of Parameter-Efficient Tuning for Image Classification
Jiacheng Ruan, Xian Gao, Suncheng Xiang, Mingye Xie, Ting Liu, Yuzhuo Fu
Native Robustness Classification Code Adversarial Sample Parameter Efficient Tuning Learnable Parameter Perturbation Attack Natural Robustness Research

October 7, 2024

R-Bench: Are your Large Multimodal Model Robust to Real-world Corruptions?
Chunyi Li, Jianbo Zhang, Zicheng Zhang, Haoning Wu, Yuan Tian, Wei Sun, Guo Lu, Xiaohong Liu, Xiongkuo Min, Weisi Lin, Guangtao Zhai
Large Multimodal Model Vision Task Image Datasets MT Bench Common Corruption Natural Robustness Research

May 3, 2024

From Attack to Defense: Insights into Deep Learning Security Measures in Black-Box Settings
Firuz Juraev, Mohammed Abuhamad, Eric Chan-Tin, George K. Thiruvathukal, Tamer Abuhmed
Adversarial Attack Native Robustness Black Box DCU Insight AQ Model Robustness Natural Robustness Research

December 16, 2023

Rethinking Robustness of Model Attributions
Sandesh Kamath, Sankalp Mittal, Amit Deshpande, Vineeth N Balasubramanian
Feature Attribution Attribution Method Model Attribution Attribution Robustness Natural Robustness Research

August 13, 2023

Understanding the robustness difference between stochastic gradient descent and adaptive gradient methods
Avery Ma, Yangchen Pan, Amir-massoud Farahmand
Gradient Descent Stochastic Gradient Descent Stochastic Way Better Robustness Adaptive Gradient Method Enhanced Robustness Natural Robustness Research

May 18, 2023

Quantifying the robustness of deep multispectral segmentation models against natural perturbations and data poisoning
Elise Bishoff, Charles Godfrey, Myles McKay, Eleanor Byler
Adversarial Attack Native Robustness Adversarial Robustness Additive Perturbation Data Poisoning Multispectral Model Intrinsic Robustness Natural Robustness Research

October 14, 2022

Pretrained Transformers Do not Always Improve Robustness
Swaroop Mishra, Bhavdeep Singh Sachdeva, Chitta Baral
Native Robustness Pre Trained Transformer Robust Representation Natural Robustness Research