Conferences and supporting programme
Secure Update of Artificial Intelligence Applications Used in an Autonomous Driving
Many applications are developed adopting intelligent hardware and algorithms using the same methodology that the biological structures are using to solve real life problems; this approach involves a lot of strategies and implementations, based on Neural Networks, Genetic Algorithm, Deep Learning, and other forms of artificial intelligence. One of the main benefits of artificial intelligence is the inherent ability to solve problems in a very short time compared to alternative implementations, while guaranteeing the robustness and the integrity of the solution - including management against unauthorized changes. Ensuring protection against unauthorized changes requires that the updates of contents (i.e. datasets) are accepted only if there is a trust between the sender and the receiver. The application of artificial intelligence in conjunction with system level security are the cornerstones of the enabling elements to realize autonomous driving. In this paper, we are focusing on these specific systems and implementations. In the first section, we provide an overview of the Artificial Intelligence concept and on the way how the intelligent systems are trained. Training is used to learn how to solve a new problem not yet submitted. The learning capability is implemented using datasets to allow the hardware to 'think' independently, so to solve not-known events based on the previous knowledge. In the second part of this talk, we will introduce the principal traits of systems that are being affected by malicious events, including attempts to corrupt the system behavior through modification of the training matrix (that is the dataset of the trained Artificial Intelligence system). Next, we will provide an overview of cryptography methodology in order to enforce the security of the system according to one of the modern usages of the cryptography. In this modern acception, cryptography is used not to hide information but to ensure the: (1) authenticity of the dataset; (2) integrity of the dataset (3) non-repudiation of the dataset and, eventually (4) avoiding the replay-attack. The fourth section of this talk introduces a cryptographic methodology that ensures that changes to the matrix content itself will be always done by an authorized entity, so a hacker can't change the functionality or the manner that Artificial Intelligence solves the issues. Finally, we will proprose a possible storage device, which implements a cryptographic engine to ensure the Secure Update of Artificial Intelligence Applications used in an Autonomous driving. In conclusion, we described a modern usage of cryptography applied to the Artificial Intellingence field and now adopted in several segments like the Autonomous driving.
--- Date: 01.03.2018 Time: 2:30 PM - 3:00 PM Location: Conference Counter NCC Ost