Seminár informačnej bezpečnosti na FIIT STU.
11. júna 2021 o 17.00 hod.
Registračný formulár: https://forms.gle/q4XKzMbg9qc4SSX37
Registrácia je otvorená do 10. júna, 23.59 hod.
Cryptography provides algorithms that enable secure communication in theory. However, in real world, these algorithms have to be implemented on real devices. To evaluate the security level of cryptographic implementations, it is necessary to include a physical security assessment.
In this talk, we will first give a brief introduction to symmetric ciphers and different physical attack methods. The working principles of one type of physical attack, EM/power analysis, and its countermeasures will be discussed in more detail. In recent years, the application of AI to EM/power analysis is gaining more attention. One line of work focuses on proposing a generic neural network structure for such attacks on various datasets with different countermeasures. An example of such a work will be presented.
On the other hand, decision-making tasks carried out by the usage of neural networks are successfully taking over in many areas, including those that are security-critical, such as healthcare, transportation, smart grids, where intentional and unintentional failures can be disastrous. Neural network implementations rely on hardware platforms (e.g., FPGAs, GPUs, and microcontrollers) to accelerate the computations. As will be shown in the first part of the talk, these physical systems are vulnerable to physical attacks. Therefore, there is a necessity to evaluate the potential attacks that can target neural networks on the hardware level. In the last part of the talk, we will present various physical attacks on AI implementations.
Xiaolu Hou is currently an assistant professor at FIIT, STU. She obtained her Ph.D. degree in Mathematics from Nanyang Technological University, Singapore in May 2017. Her research focus is on fault injection and side-channel attacks. She also has research experience in security of neural networks, location privacy, multiparty computation, and differential privacy. With a wide range of research interests, she has published her work at top venues within various fields, ranging from mathematics to computer security.
Seminár informačnej bezpečnosti na FIIT STU v Bratislave vedú Martin Mihalovič a Ing. Alexander Valach. Garant seminára je doc. Ladislav Hudec.