The integration of autonomous cyber-physical systems (CPS) in society that interface with humans necessitates assurances for safety, security, and privacy. Traditional CPS research thrusts in this space have typically focused on closed-loop, deterministic models with relatively low-dimensional physics. With the artificial intelligence renaissance, deep learning models have enabled the utility of large amounts of data stemming from heterogeneous, distributed, and cyber-physical Internet-of-Things (IoT) networks. We are witnessing the emergence of performant cyber-physical systems whose interactions are poorly understood and rapidly evolving despite widespread adoption. My recent research explores how a semantic understanding of a deep learning model’s environment can be leveraged to not only provide guarantees but also to enhance the reasoning power of a deep learning model. Neural-symbolic approaches that combine human logic with deep learning lie at the frontier of human-machine teaming in distributed and heterogeneous IoT environments. My research aims to answer the following questions: 1) how can we design neural-symbolic frameworks that are semantically conscious of their subsuming cyber-physical systems? 2) In distributed and heterogeneous IoT environments enabled with such neural-symbolic frameworks, what are the correct programming abstractions that need to be exposed to developers? 3) How can we defend against collateral safety, security, and privacy threats that will subsequently be exposed by semantically aware, sensor-rich, adaptive, and distributed heterogeneous IoT environments? This talk will provide an overview of my research with an emphasis on the latter question of safety, security, and privacy threats in this space.
Luis Garcia joined USC ISI's Networking and Cybersecurity Division as a Research Computer Scientist in June. He was previously a Postdoctoral Scholar in the Networked and Embedded Systems Laboratory (NESL) in the University of California, Los Angeles (UCLA) Electrical and Computer Engineering Department since 2018. His research interests include the safety and security of learning-enabled cyber-physical systems, malware analysis and reverse engineering, industrial control system security and verification, as well as broad interests in novel applications of machine learning. He obtained his Ph.D. in Computer Engineering with a Cyber Security track working on the safety and security of cyber-physical industrial control systems at Rutgers University in 2018.