Search

Recent innovations in machine learning enjoy a remarkable rate of adoption across a broad spectrum of applications, including cyber-security. While previous chapters study the application of machine learning solutions to cyber-security, in this chapter we present adversarial machine learning: a field of study concerned with the security of machine learning algorithms when faced with attackers. Likewise, adversarial machine learning enjoys remarkable interest from the community, with a large body of works that either propose attacks against machine learning algorithms, or defenses against adversarial attacks. In particular, adversarial attacks have been mounted in almost all applications of machine learning. Here, we aim to systematize adversarial machine learning, with a pragmatic focus on common computer security applications. Without assuming a strong background in machine learning, we also introduce the basic building blocks and fundamental properties of adversarial machine learning. This study is therefore accessible both to a security audience without in-depth knowledge of machine learning and to a machine learning audience.

We explore how data modification can enhance privacy by examining the connection between data modification and machine learning. Specifically, machine learning “meets” data modification in two ways. First, data modification can protect the data that is used to train machine learning models focusing it on the intended use and inhibiting unwanted inference. Second, machine learning can provide new ways of creating modified data. In this chapter, we discuss data modification approaches, applied during data pre-processing, that are suited for online data sharing scenarios. Specifically, we define two scenarios “User data sharing” and “Data set sharing” and describe the threat models associated with each scenario and related privacy threats. We then survey the landscape of privacy-enhancing data modification techniques that can be used to counter these threats. The picture that emerges is that data modification approaches hold promise to enhance privacy, and can be used alongside of conventional cryptographic approaches. We close with an outlook on future directions focusing on new types of data, the relationship among privacy, and the importance of taking an interdisciplinary approach to data modification for privacy enhancement.

This chapter provides an overview of recent applications of deep learning to profiled side-channel analysis (SCA). The advent of deep neural networks (mainly multiple layer perceptrons and convolutional neural networks) as a learning algorithm for profiled SCA opened several new directions and possibilities to explore the occurrence of side-channel leakages from different categories of systems. This is particularly important for designers to verify to what extent an adversary can extract sensitive information when possessing state-of-the-art attack methods. Deep learning is a fast-evolving technology that provides several advantages in profiled SCA and we summarize what are the main directions and results obtained by the research community.

ispartof: pages:103-122 ispartof: SMART CARD RESEARCH AND ADVANCED APPLICATIONS, CARDIS 2022 vol:13820 pages:103-122 ispartof: 21st International Conference on Smart Card Research and Advanced Applications (CARDIS) location:ENGLAND, Birmingham date:7 Nov - 9 Nov 2022 status: published

Item does not contain fulltext In this chapter, we are considering the physical security of Machine Learning (ML) implementations on Edge Devices. We list the state-of-the-art known physical attacks, with the main attack objectives to reverse engineer and misclassify ML models. These attacks have been reported for different target platforms with the usage of both passive and active attacks. The presented works highlight the potential threat of stealing an intellectual property or confidential model trained with private data, and also the possibility to tamper with the device during the execution to cause misclassification. We also discus possible countermeasures to mitigate such attacks.

Item does not contain fulltext We explore how data modification can enhance privacy by examining the connection between data modification and machine learning. Specifically, machine learning “meets” data modification in two ways. First, data modification can protect the data that is used to train machine learning models focusing it on the intended use and inhibiting unwanted inference. Second, machine learning can provide new ways of creating modified data. In this chapter, we discuss data modification approaches, applied during data pre-processing, that are suited for online data sharing scenarios. Specifically, we define two scenarios “User data sharing” and “Data set sharing” and describe the threat models associated with each scenario and related privacy threats. We then survey the landscape of privacy-enhancing data modification techniques that can be used to counter these threats. The picture that emerges is that data modification approaches hold promise to enhance privacy, and can be used alongside of conventional cryptographic approaches. We close with an outlook on future directions focusing on new types of data, the relationship among privacy, and the importance of taking an interdisciplinary approach to data modification for privacy enhancement.

Item does not contain fulltext This chapter provides an overview of recent applications of deep learning to profiled side-channel analysis (SCA). The advent of deep neural networks (mainly multiple layer perceptrons and convolutional neural networks) as a learning algorithm for profiled SCA opened several new directions and possibilities to explore the occurrence of side-channel leakages from different categories of systems. This is particularly important for designers to verify to what extent an adversary can extract sensitive information when possessing state-of-the-art attack methods. Deep learning is a fast-evolving technology that provides several advantages in profiled SCA and we summarize what are the main directions and results obtained by the research community.

Contains fulltext : 250777.pdf (Publisher’s version ) (Closed access) Recent innovations in machine learning enjoy a remarkable rate of adoption across a broad spectrum of applications, including cyber-security. While previous chapters study the application of machine learning solutions to cyber-security, in this chapter we present adversarial machine learning: a field of study concerned with the security of machine learning algorithms when faced with attackers. Likewise, adversarial machine learning enjoys remarkable interest from the community, with a large body of works that either propose attacks against machine learning algorithms, or defenses against adversarial attacks. In particular, adversarial attacks have been mounted in almost all applications of machine learning. Here, we aim to systematize adversarial machine learning, with a pragmatic focus on common computer security applications. Without assuming a strong background in machine learning, we also introduce the basic building blocks and fundamental properties of adversarial machine learning. This study is therefore accessible both to a security audience without in-depth knowledge of machine learning and to a machine learning audience.