Native Robustness
Native robustness in machine learning focuses on developing models inherently resistant to various forms of input perturbations, including adversarial attacks and noisy data, without relying solely on post-hoc defenses. Current research emphasizes techniques like ensemble methods, reprogramming existing models, and modifying training procedures (e.g., using different learning rates for specific model layers or incorporating regularization methods) to improve robustness across diverse model architectures, including convolutional neural networks, vision transformers, and large language models. This field is crucial for deploying reliable AI systems in safety-critical applications, such as healthcare and autonomous driving, where model resilience to unexpected inputs is paramount.
Papers
Comparing Prior and Learned Time Representations in Transformer Models of Timeseries
Natalia Koliou, Tatiana Boura, Stasinos Konstantopoulos, George Meramveliotakis, George Kosmadakis
Perfecting Imperfect Physical Neural Networks with Transferable Robustness using Sharpness-Aware Training
Tengji Xu, Zeyu Luo, Shaojie Liu, Li Fan, Qiarong Xiao, Benshan Wang, Dongliang Wang, Chaoran Huang
Uncertainty-based Offline Variational Bayesian Reinforcement Learning for Robustness under Diverse Data Corruptions
Rui Yang, Jie Wang, Guoping Wu, Bin Li
Cityscape-Adverse: Benchmarking Robustness of Semantic Segmentation with Realistic Scene Modifications via Diffusion-Based Image Editing
Naufal Suryanto, Andro Aprila Adiputra, Ahmada Yusril Kadiptya, Thi-Thu-Huong Le, Derry Pratama, Yongsu Kim, Howon Kim
Certified Robustness for Deep Equilibrium Models via Serialized Random Smoothing
Weizhi Gao, Zhichao Hou, Han Xu, Xiaorui Liu