Sparse Perturbation

Sparse perturbation research focuses on understanding and mitigating the impact of small, targeted changes to data or model parameters. Current efforts concentrate on developing efficient algorithms for generating and defending against these sparse adversarial attacks, particularly within the context of image classification and language models, often employing techniques like projected gradient descent and sharpness-aware minimization with sparsified perturbations. This work is crucial for improving the robustness and security of machine learning systems, addressing vulnerabilities to malicious manipulations and enhancing the reliability of model explanations in various applications.

Papers

May 23, 2024

Certified Robustness against Sparse Adversarial Perturbations via Data Localization
Ambar Pal, René Vidal, Jeremias Sulam
Native Robustness Adversarial Robustness Adversarial Perturbation Localization Focus Robustness Guarantee Robust Classifier Sparse Attack Sparse Perturbation

May 8, 2024

Towards Efficient Training and Evaluation of Robust Models against $l_0$ Bounded Adversarial Perturbations
Xuyang Zhong, Yixiao Huang, Chen Liu
Global Evaluation Adversarial Perturbation Efficient Training Robust Model Sparse Attack Sparse Perturbation Box Attack

April 8, 2024

BruSLeAttack: A Query-Efficient Score-Based Black-Box Sparse Adversarial Attack
Viet Quoc Vo, Ehsan Abbasnejad, Damith C. Ranasinghe
Adversarial Example Adversarial Sample QuEry Based Attack Sparse Attack Sparse Perturbation

January 16, 2024

Explaining Time Series via Contrastive and Locally Sparse Perturbations
Zichuan Liu, Yingying Zhang, Tianchun Wang, Zefan Wang, Dongsheng Luo, Mengnan Du, Min Wu, Yi Wang, Chunlin Chen, Lunting Fan, Qingsong Wen
Contrastive Learning Time Series Multivariate Time Series Contrastive Example Sparse Model Temporal Pattern Sparse Perturbation

June 30, 2023

Systematic Investigation of Sparse Perturbed Sharpness-Aware Minimization Optimizer
Peng Mi, Li Shen, Tianhe Ren, Yiyi Zhou, Tianshuo Xu, Xiaoshuai Sun, Tongliang Liu, Rongrong Ji, Dacheng Tao
Many Sparse Sharpness Aware Minimization Systematic Study Sparse Mask Sparse Perturbation

October 11, 2022

June 2, 2022

Weakly Supervised Representation Learning with Sparse Perturbations
Kartik Ahuja, Jason Hartford, Yoshua Bengio
Latent Variable Weak Supervision Weakly Supervised Sparse Perturbation

February 15, 2022

Federated Learning with Sparsified Model Perturbation: Improving Accuracy under Client-Level Differential Privacy
Rui Hu, Yanmin Gong, Yuanxiong Guo
Differential Privacy Privacy Analysis Differentially Private Federated Learning Sparse Perturbation Client Level Differential Privacy

Sparse Perturbation

Papers

Certified Robustness against Sparse Adversarial Perturbations via Data Localization

Towards Efficient Training and Evaluation of Robust Models against $l_0$ Bounded Adversarial Perturbations

BruSLeAttack: A Query-Efficient Score-Based Black-Box Sparse Adversarial Attack

Explaining Time Series via Contrastive and Locally Sparse Perturbations

Systematic Investigation of Sparse Perturbed Sharpness-Aware Minimization Optimizer

Improving Sharpness-Aware Minimization with Fisher Mask for Better Generalization on Language Models

Make Sharpness-Aware Minimization Stronger: A Sparsified Perturbation Approach

Weakly Supervised Representation Learning with Sparse Perturbations

Federated Learning with Sparsified Model Perturbation: Improving Accuracy under Client-Level Differential Privacy