CSE Doctoral Student Seminar: Yehan Ma and Tara Salaman

Oct 21, 2016
12:30 p.m.
2 p.m.
Lopata Hall, Room 101

"Towards a Holistic Management of Wireless Control System"

Yehan Ma
Advisor: Chenyang Lu

Wireless sensor-actuator networks (WSANs) are gaining adoption for industrial process automation to lower deployment and maintenance costs. However, traditional wireless control systems do not support run-time coordination between the plant controller and the network manager. To enhance the resiliency and efficiency of wireless control we propose a holistic management framework for wireless control systems that closes the loop between control and the network. The key to our framework is a holistic controller that not only provides input signals to physical plants, but also reconfigures the WSAN to maintain desired performance levels based on control needs. We then develop a concrete holistic management approach that dynamically reconfigures the number of retransmissions in a WSAN based on online observations of relevant physical and cyber variables. We implement our holistic management system in the Wireless Cyber-Physical Simulator (WCPS). A case study based on a 5-state system and 16-node WSAN demonstrates that our holistic control scheme effectively enhances the system resilience against both wireless interferences and physical disturbances, while reducing the number of transmissions in the network.

"Machine Learning for Network Security"

Tara Salaman
Advisor: Raj Jain

Recently, machine learning techniques have attracted lots of attention in the research community to build intrusion detection systems (IDS) that can detect anomalous flows in a network traffic. However, most of the research does not distinguish between different types of traffic which is a necessary step to build appropriate countermeasures against particular attacks. Applying machine learning to a network security dataset, we show that even if the attack detection accuracy is good, categorization can have a low accuracy using classical techniques. We improve the classification accuracy by merging a few categories and changing the classification strategy.