Worst Case Robustness

Worst-case robustness in machine learning focuses on designing models that perform well even under the most adverse conditions, such as adversarial attacks or significant data distribution shifts. Current research emphasizes developing methods to certify robustness (proving a model's resilience to perturbations), analyzing the impact of model architecture (e.g., neural networks, particularly graph neural networks) and training techniques (e.g., adversarial training, importance weighting, jittering) on worst-case performance, and exploring efficient algorithms for evaluating and improving robustness. This research is crucial for deploying reliable machine learning systems in safety-critical applications where unexpected inputs or data variations could have severe consequences.

Papers

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

May 10, 2024

Sharp analysis of out-of-distribution error for "importance-weighted" estimators in the overparameterized regime
Kuo-Wei Lai, Vidya Muthukumar
Gaussian Mixture Model Linear Estimator Chaotic Regime Importance Weight Worst Case Robustness

February 28, 2024

A Quick Framework for Evaluating Worst Robustness of Complex Networks
Wenjun Jiang, Peiyan Li, Tianlong Fan, Ting Li, Chuan-fu Zhang, Tao Zhang, Zong-fu Luo
Native Robustness Complex Network Attack Strategy Worst Case Robustness

November 2, 2023

Upper and lower bounds for the Lipschitz constant of random neural networks
Paul Geuchen, Thomas Heindl, Dominik Stöger, Felix Voigtlaender
Adversarial Example Adversarial Perturbation Lower Bound Shallow Neural Network Lipschitz Constant Random Neural Network Worst Case Robustness

October 15, 2023

Robust Multi-Agent Reinforcement Learning by Mutual Information Regularization
Simin Li, Ruixiao Xu, Jingqiao Xiu, Yuwei Zheng, Pu Feng, Yaodong Yang, Xianglong Liu
Multi Agent Reinforcement Learning Worst Case Robustness Robust Strategy Mutual Information Regularization Action Prior

July 24, 2023

Learning Provably Robust Estimators for Inverse Problems via Jittering
Anselm Krainovic, Mahdi Soltanolkotabi, Reinhard Heckel
Inverse Problem Worst Case Robustness Model Jitter Optimal Denoising

May 1, 2023

Robustified Learning for Online Optimization with Memory Costs
Pengfei Li, Jianyi Yang, Shaolei Ren
Online Algorithm Online Optimization Downstream Forecasting Task Worst Case Robustness

February 8, 2023

WAT: Improve the Worst-class Robustness in Adversarial Training
Boqi Li, Weiwei Liu
Adversarial Attack Adversarial Example Adversarial Training Worst Case Robustness

February 6, 2023

Bitrate-Constrained DRO: Beyond Worst Case Robustness To Unknown Group Shifts
Amrith Setlur, Don Dennis, Benjamin Eysenbach, Aditi Raghunathan, Chelsea Finn, Virginia Smith, Sergey Levine
Robust Training Worst Case Robustness CVaR Objective

November 10, 2022

On Proper Learnability between Average- and Worst-case Robustness
Vinod Raman, Unique Subedi, Ambuj Tewari
Robust Loss Worst Case Robustness Average Performance Robust PAC

November 7, 2022

Max-Min Off-Policy Actor-Critic Method Focusing on Worst-Case Robustness to Model Misspecification
Takumi Tanabe, Rei Sato, Kazuto Fukuchi, Jun Sakuma, Youhei Akimoto
Reinforcement Learning Actor Critic Deep Deterministic Policy Gradient Model Misspecification Policy Algorithm Policy Actor Critic Uncertainty Set Multi Joint Worst Case Robustness

June 13, 2022

Robust Distillation for Worst-class Performance
Serena Wang, Harikrishna Narasimhan, Yichen Zhou, Sara Hooker, Michal Lukasik, Aditya Krishna Menon
Knowledge Distillation Robust Optimization Strong Learner Worst Case Robustness Worst Class

February 2, 2022

Probabilistically Robust Learning: Balancing Average- and Worst-case Performance
Alexander Robey, Luiz F. O. Chamon, George J. Pappas, Hamed Hassani
Native Robustness Adversarial Training Robust Bayesian Worst Case Robustness Probabilistic Robustness Average Performance Worst Case Performance