• Aucun résultat trouvé

An Optimal Control View of Adversarial Machine Learning

N/A
N/A
Info
Télécharger
Protected

Academic year: 2022

Partager "An Optimal Control View of Adversarial Machine Learning"

Copied!
1
0
0

Chargement.... (Voir le texte intégral maintenant)

Texte intégral

(1)

Keynote: An Optimal Control View of Adversarial Machine Learning

Xiaojin Zhu

∗

Abstract

Test-time adversarial examples, training set poisoning, re- ward shaping, etc.: these attacks as studied in adversarial ma- chine learning have one thing in common: the adversary liter- ally wants to control a machine learning system. In this talk, we will develop this connection to control theory. The result- ing view allows more clarity into adversarial learning, and opens up promising research directions.

∗X. Zhu is with the Department of Computer Science, University of Wisconsin-Madison, WI, USA. e-mail: jer- ryzhu@cs.wisc.edu

Copyright c⃝by the paper’s authors. Copying permitted for private and academic purposes. In: Joseph Collins, Prithviraj Dasgupta, Ranjeev Mittu (eds.): Proceedings of the AAAI Fall 2018 Sympo- sium on Adversary-Aware Learning Techniques and Trends in Cy- bersecurity, Arlington, VA, USA, 18-19 October, 2018, published at http://ceur-ws.org

Références

Télécharger maintenant ( PDF - 1 Page - 73.79 KB )

Documents relatifs

Learning with stochastic inputs and adversarial outputs

Although the adversarial setting allows to consider a wide range of problems by dropping any assumption about the way data are generated, in many applications it is possible to

Symmetric adversarial poisoning against deep learning

Deng, “Very deep convolutional neural network based image classification using small training sample size,” in 2015 3rd IAPR Asian Conference on Pattern Recognition (ACPR), Nov

АНАЛИЗ КОММЕРЧЕСКОГО ТЕРМИНА И ПУТИ ЕГО ПЕРЕВОДА НА МАТЕРИАЛЕ ДОКУМЕНТОВ ИЗ РУССКОГО НА ФРАНЦУЗСКЙ И АРАБСКИЙ ЯЗЫКИ

Система уровней владения РЯДО по сформировенности коммуникативных умений может быть соотнесена с системой уровней общего владения иностраными

About the inevitability of adversarial examples

Indeed, if one consider adversarial examples as points close to decision boundary, then, one can prove that adversarial examples are not only inevitable but almost everywhere..

Hybrid Stochastic-Adversarial On-line Learning

Both EStochAd and the Agnostic Online Learning (AOL) algorithm proposed in [SS08] consider the problem of binary classification with adversarial outputs, an infinite number of

Deep Conditional Adversarial Learning for Polyp Segmentation

So, motivated by the insights of GANs in the domain of medical image analysis [8], we have proposed a deep conditional adversarial learning based network for automatic

Motion Capture Synthesis with Adversarial Learning

We propose a new statistical modeling approach that we call Sequential Adversarial Auto-encoder (SAAE) for learning a synthesis model for motion sequences.. This model exploits

Smooth Adversarial Examples

Machine assessment of perceptual similarity between two images the input image and its adversarial example is arguably as difficult as the original classification task, while