Rearchitecting Classification Frameworks For Increased Robustness

2019/05/26 by Varun Chandrasekaran, Chandrasekaran, Varun, Brian Tang +9
Computer Science · #Advanced Malware Detection Techniques #Adversarial Robustness in Machine Learning #Anomaly Detection Techniques and Applications #FOS: Computer and information sciences #Machine Learning (cs.LG) #Machine Learning (stat.ML)

paper · pdf · doi:10.48550/arxiv.1905.10900

openalex publication_date 2019/05/26 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28

Abstract

While generalizing well over natural inputs, neural networks are vulnerable to adversarial inputs. Existing defenses against adversarial inputs have largely been detached from the real world. These defenses also come at a cost to accuracy. Fortunately, there are invariances of an object that are its salient features; when we break them it will necessarily change the perception of the object. We find that applying invariants to the classification task makes robustness and accuracy feasible together. Two questions follow: how to extract and model these invariances? and how to design a classification paradigm that leverages these invariances to improve the robustness accuracy trade-off? The remainder of the paper discusses solutions to the aformenetioned questions.

Citations

• Intriguing properties of neural networks
• Distilling the Knowledge in a Neural Network
• Statistical Decision Functions
• Vision meets robotics: The KITTI dataset
• Learning with a Strong Adversary
• Transferability in Machine Learning: from Phenomena to Black-Box Attacks using Adversarial Samples
• Neural Architecture Search with Reinforcement Learning
• Axiomatic Attribution for Deep Networks
• Comparing deep neural networks against humans: object recognition when\n the signal gets weaker
• Robust Physical-World Attacks on Deep Learning Models
• Certifying Some Distributional Robustness with Principled Adversarial\n Training
• Distilling a Neural Network Into a Soft Decision Tree
• Deep k-Nearest Neighbors: Towards Confident, Interpretable and Robust Deep Learning
• YOLOv3: An Incremental Improvement
• Physical Adversarial Examples for Object Detectors
• Generating 3D Adversarial Point Clouds
• Designing Adversarially Resilient Classifiers using Resilient Feature Engineering
• Computational Limitations in Robust Classification and Win-Win Results
• On Evaluating Adversarial Robustness
• Improving Robustness of ML Classifiers against Realizable Evasion Attacks Using Conserved Features
• Exploiting Excessive Invariance caused by Norm-Bounded Adversarial\n Robustness
• Adversarial Examples Are Not Bugs, They Are Features
• You Only Propagate Once: Accelerating Adversarial Training via Maximal\n Principle
• Invariant Risk Minimization

Related

• Improve Adversarial Robustness via Weight Penalization on Classification Layer
• Improving the Adversarial Robustness and Interpretability of Deep Neural\n Networks by Regularizing their Input Gradients
• Robustness from Simple Classifiers
• Adversarial Robustness as a Prior for Learned Representations
• Understanding Misclassifications by Attributes