Artificial Intelligence (AI)

Overview

While the term “Artificial Intelligence (AI)” has grown to encompass a wide variety of meanings, it is based on the idea that human intelligence can be emulated by machines via learning, perceiving and reasoning. AI can be used for a variety of tasks such as character recognition, process-guidance, gaming, drug manufacturing etc. Machine learning (ML) is a branch of AI tasked with learning through experience via exposure to large sets of data and drawing patterns and correlations among various factors.

Our group uses data-mining techniques to extract patterns that serve as “signatures” for physical systems and attempts to detect unauthorized intrusion that violate these signatures using ML. Other applications of AI include reinforcement learning for additive manufacturing.

Relevant work

Validation of Covert Cognizance Active Defenses

This work by Arvind Sundaram assesses the effectiveness of the so-called Covert Cognizance (C²) paradigm using state-of-the-art generative adversarial nets (GANs). The GAN model is trained on real data while the discriminator is used to differentiate C²-embedded data from real data. It is observed that the GAN is unable to do so better than random, thus validating the effectiveness of the C² paradigm.

Effectiveness of Model-Based Defenses for Digitally Controlled Industrial Systems: Nuclear Reactor Case Study

This work by Yeni Li derives model-based signatures from complex models using ROM techniques. Using ML, it analyzes their effectiveness in detecting cybersecurity attacks assuming various levels of attacker knowledge. Model-based signatures have been touted as essential defenses to detect intrusion over the past decade and serve as an additional layer of security for industrial control systems (ICS).