Open AccessPosted Content10.20944/PREPRINTS202009.0184.V1

Perspectives on Adversarial Classification

- 08 Sep 2020

TL;DR: Alternative perspectives on AC based on adversarial risk analysis are presented, including a classical game-theoretic framework, which requires unrealistic common knowledge conditions untenable in the security settings typical of the AML realm.

Abstract: Adversarial classification (AC) is a major subfield within the increasingly important domain of adversarial machine learning (AML). So far, most approaches to AC have followed a classical game-theoretic framework. This requires unrealistic common knowledge conditions untenable in the security settings typical of the AML realm. After reviewing such approaches, we present alternative perspectives on AC based on adversarial risk analysis.

AI Agents for this Paper

Find similar papers on Google Scholar, PubMed and Arxiv
Write a critical review of this paper
Analyze citations of this paper to find unaddressed research gaps

Most frequently asked questions

1. What contributions have the authors mentioned in the paper "Perspectives on adversarial classification" ?

After reviewing such approaches, the authors present alternative perspectives on AC based on adversarial risk analysis.

2. What are the future works mentioned in the paper "Perspectives on adversarial classification" ?

To this end, strategies to more efficiently explore the highly complex, multimodal posterior distributions of neural models constitute another line of further work.

Figure 5. Ten MNIST examples (top) and their perturbations (bottom). Predicted class shown for each example.

Table 1. Accuracy comparison (with precision) of four classifiers on clean (untainted) and attacked (tainted) data.

Figure 6. Robustness of a deep network for MNIST under three defence mechanisms (none, AT, ARA). (a) depicts the security evaluation curves under the FGSM attack. (b) depicts the security evaluation curves under the PGD attack.

Table 2. Accuracy comparison (with precision) of four classifiers on attacked data, with and without ARA-enhanced defence.

Citations

•Journal Article•10.1002/WICS.1530

Adversarial risk analysis: An overview

David Banks, +4 more

- 14 Sep 2020

- Wiley Interdisciplinary Reviews: Computa...

TL;DR: Adversarial risk analysis (ARA) as discussed by the authors is a relatively new area of research that informs decision-making when facing intelligent opponents and uncertain outcomes, and it is a decision-theoretic alternative to game theory.

...read moreread less

Journal Article•10.1016/j.ejor.2022.06.042

Augmented probability simulation methods for sequential games

Tahir Ekin, +3 more

- 01 Jun 2022

- European Journal of Operational Research

TL;DR: In this paper , a robust framework with computational algorithms to support decision makers in sequential games is presented, which includes methods to solve games with complete information, assess the robustness of such solutions and, finally, approximate adversarial risk analysis solutions when lacking complete information.

...read moreread less

•Posted Content•10.20944/PREPRINTS202009.0184.V1

Perspectives on Adversarial Classification

David Ríos Insua, +2 more

- 08 Sep 2020

...read moreread less

Book Chapter•10.1007/978-3-030-99772-4_4

Game Theoretical Adversarial Deep Learning

26 Aug 2022

•Posted Content

Adversarial attacks against Bayesian forecasting dynamic models.

Roi Naveiro

- 20 Oct 2021

- arXiv: Machine Learning

TL;DR: In this article, a decision analysis based attack strategy was proposed to attack Bayesian forecasting dynamic models, which could be used against time series forecasting systems, such as time series regression and classification systems.

...read moreread less

References

Book Chapter•10.1007/978-3-658-40442-0_9

Generative Adversarial Nets

Tania Leach

- 01 Jan 2023

18.7K

•Posted Content

Explaining and Harnessing Adversarial Examples

Ian Goodfellow, +2 more

- 20 Dec 2014

- arXiv: Machine Learning

TL;DR: The authors argue that the primary cause of neural networks' vulnerability to adversarial perturbation is their linear nature, which is supported by new quantitative results while giving the first explanation of the most intriguing fact about adversarial examples: their generalization across architectures and training sets.

...read moreread less

15.9K

Probabilistic Outputs for Support vector Machines and Comparisons to Regularized Likelihood Methods

John Platt

- 01 Jan 1999

TL;DR: The output of a lassi(cid:12)er should be a alibrated posterior probability to enable post-pro essing and a method to train a kernel lassi with a logit link and a regularized maximum likelihood is proposed.

...read moreread less

6.1K

•Posted Content

Intriguing properties of neural networks

Christian Szegedy, +7 more

- 21 Dec 2013

- arXiv: Computer Vision and Pattern Recog...

TL;DR: This article showed that deep neural networks learn input-output mappings that are fairly discontinuous to a significant extend, which suggests that it is the space, rather than individual units, that contains of the semantic information in the high layers of neural networks.

...read moreread less

5.8K

An empirical study of the naive Bayes classifier

Irina Rish

- 01 Jan 2001

TL;DR: This work analyzes the impact of the distribution entropy on the classification error, showing that low-entropy feature distributions yield good performance of naive Bayes and demonstrates that naive Baye works well for certain nearlyfunctional feature dependencies.

...read moreread less

2.6K

...

Expand

Perspectives on Adversarial Classification

Chat with Paper

AI Agents for this Paper

Most frequently asked questions

1. What contributions have the authors mentioned in the paper "Perspectives on adversarial classification" ?

2. What are the future works mentioned in the paper "Perspectives on adversarial classification" ?

Figures

Citations

Adversarial risk analysis: An overview

Augmented probability simulation methods for sequential games

Perspectives on Adversarial Classification

Game Theoretical Adversarial Deep Learning

Adversarial attacks against Bayesian forecasting dynamic models.

References

Generative Adversarial Nets

Explaining and Harnessing Adversarial Examples

Probabilistic Outputs for Support vector Machines and Comparisons to Regularized Likelihood Methods

Intriguing properties of neural networks

An empirical study of the naive Bayes classifier

Related Papers (5)

Perspectives on Adversarial Classification

Robust Assessment of Real-World Adversarial Examples

Benchmarking Adversarial Robustness on Image Classification

DeepEC: Adversarial attacks against graph structure prediction models

Identifying Security Evaluation of PatternClassifiers Under attack