A Fundamental Performance Limitation for Adversarial Classification
- 10 June 2019
- journal article
- research article
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Control Systems Letters
- Vol. 4 (1) , 169-174
Abstract
Despite the widespread use of machine learning algorithms to solve problems of technological, economic, and social relevance, provable guarantees on the performance of these data-driven algorithms are critically lacking, especially when the data originates from unreliable sources and is transmitted over unprotected and easily accessible channels. In this letter, we take an important step to bridge this gap and formally show that, in a quest to optimize their accuracy, binary classification algorithms-including those based on machine-learning techniques-inevitably become more sensitive to adversarial manipulation of the data. Further, numerical evidence suggests that the accuracy-sensitivity tradeoff depends solely on the statistics of the data, and cannot be improved by tuning the algorithms or increasing their complexity.Keywords
All Related Versions
Funding Information
- ARO (ARO-71603NSYIP)
- ONR (ONR-N00014-19-1-2264)
This publication has 8 references indexed in Scilit:
- A Tour of Reinforcement Learning: The View from Continuous ControlAnnual Review of Control, Robotics, and Autonomous Systems, 2019
- Adversarial Regression for Detecting Attacks in Cyber-Physical Systems2018
- Deep learning feature extraction for target recognition and classification in underwater sonar images2017
- DeepFool: A Simple and Accurate Method to Fool Deep Neural Networks2016
- The Limitations of Deep Learning in Adversarial Settings2016
- Reinforcement Learning and Feedback Control: Using Natural Decision Methods to Design Optimal Adaptive ControllersIEEE Control Systems, 2012
- Recent Advances of Large-Scale Linear ClassificationProceedings of the IEEE, 2012
- Adversarial learning2005