Robustness Guarantee

Robustness guarantees in machine learning aim to ensure that models maintain reliable performance even when faced with uncertainties, such as noisy data, adversarial attacks, or model parameter changes. Current research focuses on developing methods to certify robustness, often employing techniques like randomized smoothing, Lipschitz constraints, and interval abstractions, applied to various model architectures including neural networks and Bayesian networks. These advancements are crucial for deploying machine learning models in safety-critical applications, where reliable performance under uncertainty is paramount, and for improving the trustworthiness of AI systems more broadly.

Papers

September 28, 2023

Adversarial Examples Might be Avoidable: The Role of Data Concentration in Adversarial Robustness
Ambar Pal, Jeremias Sulam, René Vidal
Adversarial Example Adversarial Robustness Integral Role Robustness Guarantee Robust Classifier High Quality Subset Text Consolidation

September 22, 2023

Provably Robust and Plausible Counterfactual Explanations for Neural Networks via Robust Optimisation
Junqi Jiang, Jianglin Lan, Francesco Leofante, Antonio Rago, Francesca Toni
Neural Network Native Robustness Counterfactual Explanation Robustness Guarantee Robust Optimisation

July 24, 2023

Adaptive Certified Training: Towards Better Accuracy-Robustness Tradeoffs
Zhakshylyk Nurlanov, Frank R. Schmidt, Florian Bernard
Native Robustness Robustness Guarantee Accuracy Robustness Adaptive Training TinyImagenet Benchmark

June 16, 2023

Learning-Augmented Decentralized Online Convex Optimization in Networks
Pengfei Li, Jianyi Yang, Adam Wierman, Shaolei Ren
Network Programming Online Convex Optimization Robustness Guarantee Learning Augmented

June 13, 2023

Theoretical Foundations of Adversarially Robust Learning
Omar Montasser
Machine Learning Adversarial Example Robust Learning Robustness Guarantee Theoretical Foundation Robust Adversarial Learning

June 12, 2023

How robust accuracy suffers from certified training with convex relaxations
Piersilvio De Bartolomeis, Jacob Clarysse, Amartya Sanyal, Fanny Yang
Adversarial Attack Training Data Adversarial Training Robust Training Robustness Guarantee Convex Relaxation Threat Model

June 7, 2023

Label Shift Quantification with Robustness Guarantees via Distribution Feature Matching
Bastien Dussap, Gilles Blanchard, Badr-Eddine Chérief-Abdellatif
Label Distribution Robustness Guarantee Label Shift Label Distribution Shift Generalized Label Shift Feature Distribution Matching

May 31, 2023

May 26, 2023

A Tale of Two Approximations: Tightening Over-Approximation for DNN Robustness Verification via Under-Approximation
Zhiyi Xue, Si Liu, Zhaodi Zhang, Yiting Wu, Min Zhang
Average Approximation Robustness Guarantee Cautionary TALE Approximation Property Tight Approximation

March 20, 2023

Lipschitz-bounded 1D convolutional neural networks using the Cayley transform and the controllability Gramian
Patricia Pauli, Ruigang Wang, Ian R. Manchester, Frank Allgöwer
Convolutional Neural Network Convolutional Layer Robustness Guarantee Optimization Layer Gram Matrix

February 22, 2023

Provable Robustness Against a Union of $\ell_0$ Adversarial Attacks
Zayd Hammoudeh, Daniel Lowd
Adversarial Attack Feature Aggregation Robustness Guarantee Robustness Analysis

January 31, 2023

January 30, 2023

Risk-Averse Model Uncertainty for Distributionally Robust Safe Reinforcement Learning
James Queeney, Mouhacine Benosman
Deep Reinforcement Learning Safe Reinforcement Learning Robustness Guarantee Uncertain Environment

January 26, 2023

Certified Interpretability Robustness for Class Activation Mapping
Alex Gu, Tsui-Wei Weng, Pin-Yu Chen, Sijia Liu, Luca Daniel
Native Robustness Class Activation Robustness Guarantee Robust Interpretation

January 7, 2023

REaaS: Enabling Adversarially Robust Downstream Classifiers via Robust Encoder as a Service
Wenjie Qu, Jinyuan Jia, Neil Zhenqiang Gong
Adversarial Example Service Provider Robustness Guarantee Downstream Classification Robust Encoders

December 13, 2022

Statistical Safety and Robustness Guarantees for Feedback Motion Planning of Unknown Underactuated Stochastic Systems
Craig Knuth, Glen Chou, Jamie Reese, Joe Moore
Robustness Guarantee Learned Dynamic Underactuated System Probabilistic Safety Goal Reachability Feedback Motion Planning

November 21, 2022

DualApp: Tight Over-Approximation for Neural Network Robustness Verification via Under-Approximation
Yiting Wu, Zhaodi Zhang, Zhiyi Xue, Si Liu, Min Zhang
Robustness Guarantee Approximation Property Tight Approximation

November 17, 2022

VeriCompress: A Tool to Streamline the Synthesis of Verified Robust Compressed Neural Networks from Scratch
Sawinder Kaur, Yi Xiao, Asif Salekin
Critical Synthesis BRIO Tool Scratch Project Neural Network Compression Robustness Guarantee Safety Critical Application

Robustness Guarantee

Papers

Adversarial Examples Might be Avoidable: The Role of Data Concentration in Adversarial Robustness

Provably Robust and Plausible Counterfactual Explanations for Neural Networks via Robust Optimisation

Adaptive Certified Training: Towards Better Accuracy-Robustness Tradeoffs

Learning-Augmented Decentralized Online Convex Optimization in Networks

Theoretical Foundations of Adversarially Robust Learning

How robust accuracy suffers from certified training with convex relaxations

Label Shift Quantification with Robustness Guarantees via Distribution Feature Matching

Learning for Edge-Weighted Online Bipartite Matching with Robustness Guarantees

Incremental Randomized Smoothing Certification

A Tale of Two Approximations: Tightening Over-Approximation for DNN Robustness Verification via Under-Approximation

Lipschitz-bounded 1D convolutional neural networks using the Cayley transform and the controllability Gramian

Provable Robustness Against a Union of $\ell_0$ Adversarial Attacks

Interpreting Robustness Proofs of Deep Neural Networks

Optimal Transport Perturbations for Safe Reinforcement Learning with Robustness Guarantees

Risk-Averse Model Uncertainty for Distributionally Robust Safe Reinforcement Learning

Certified Interpretability Robustness for Class Activation Mapping

REaaS: Enabling Adversarially Robust Downstream Classifiers via Robust Encoder as a Service

Statistical Safety and Robustness Guarantees for Feedback Motion Planning of Unknown Underactuated Stochastic Systems

DualApp: Tight Over-Approximation for Neural Network Robustness Verification via Under-Approximation

VeriCompress: A Tool to Streamline the Synthesis of Verified Robust Compressed Neural Networks from Scratch