CECS Assistant Professor Research Talk

PEC 1430/40 (Food Provided)  or  https://umich.zoom.us/j/92644868178  (Passcode 377011)

Sample-efficient decision-making models using human-in-the-loop

Srijita Das, CIS

Abstract

This talk will focus on why humans are an “integral part” in building sample-efficient and safe machine learning models. It will discuss ways in which advice from humans or other external knowledge sources can be leveraged to help these models make better decisions. It will further delve into the cost-consideration that needs to be accounted for when involving humans in intelligent systems. Lastly, it will discuss a few open research directions that she is currently pursuing with scope for future collaboration.

Biography
Srijita Das is currently an Assistant Professor at the Department of Computer and Information Sciences (CIS) in University of Michigan-Dearborn. Prior to that, she was a Postdoctoral fellow in the Intelligent Robot Learning Lab (IRL Lab) at University of Alberta, Canada. She earned her Ph.D. in Computer Science from the University of Texas at Dallas, USA. Her research interests include human-in-the-loop Reinforcement learning, Active-learning, Cost-sensitive learning and Human-machine collaboration. She has published several research works in refereed journals and proceedings in the areas of machine learning including IJCAI, AAMAS, AAAI, JAIR & TMLR. She has also served as reviewer for several international journals (JAIR, TPAMI) and conferences (AAAI, SDM, AISTATS, AAMAS).
 

Cyber-physical Security of Critical Infrastructure
 

Junho Hong, ECE
 

Abstract

The use of industrial control systems (ICS) for controlling critical components in a power grid is becoming more prevalent. Additionally, asset monitoring devices employing sensors for reporting voltage, current, temperature, and other measurements are becoming more common for power system operations. However, due to the high penetration of Information and Communications Technology (ICT), Supervisory Control And Data Acquisition (SCADA) systems are interconnected with one another, resulting in higher vulnerability with respect to cyber intrusions. Cyberattacks on critical infrastructures are evolving, and their patterns are diversifying, particularly for energy delivery systems. Cyberattacks can damage physical systems by compromising their ICT infrastructure and gaining access to the controls and monitoring of energy delivery systems, without a requiring a physical attack. In this presentation, state-of-the-art power system domain based cyberattack detection, impact analysis and mitigation methods will be discussed together with their practical implementations in various fields (e.g., digital substation). Detection is based not on conventional cyber network defense (e.g., firewall and intrusion detection system), but on the controllers assessing correctness in the context of a physical power system state, with application of physical laws and engineering principles. The result of each project has been implemented into commercial products and demonstrated at utility/national lab partners. New intelligent of cyber security design, operation, and metrics will be important for resiliency of future energy delivery systems.
 

Biography

Junho Hong is an Assistant Professor in the Department of Electrical and Computer Engineering at the University of Michigan–Dearborn. He received his Ph.D. degree in Cybersecurity of Substation Automation Systems in Electrical Engineering from Washington State University, Pullman in 2014. During 2014-2019, he worked with ABB where he provided technical project leadership and supports strategic corporate technology development/productization in the areas related to cyber-physical security for substations, power grids control and protection, renewable integration, and utility communications. He has been working on cybersecurity of energy delivery systems with the Department of Energy (DOE) as Principal Investigator (PI) and Co-PI in the areas of substation, microgrid, HVDC, FACTS, and high power EV charger. He has 13 US patents and serves in Cigre WG D2.50-Electric power utilities’ cybersecurity for contingency operations, C4.64 on the application of real-time digital simulation in power systems, B5.77 requirements for Information Technologies (IT) and Operational Technology (OT) managed of Protection, Automation and Control Systems (PACS) and UCAIug IEC61850.