Arun Ganesan
is a Ph.D. candidate at the University of Michigan. He works under Dr. Kang Shin as part of the Real Time Computing Lab. He earned B.S. in Computer Science and Mathematics, and an M.S. in Computer Science also at the University of Michigan.

His current research is on vehicular cyber-security. He builds tools to better understand normal and anomalous behavior from large-scale vehicular data.

Previously, he worked on gesture recognition systems using smartwatches, auditory sensing, and the Microsoft Kinect. Prior to that, he studied the risk of mobile WiFi-based botnets and how to defend against them.
Vehicular Cybersecurity
Vehicular Anomaly Detection

As we enter the age of intelligent and self-driving cars, cyber-physical security of our vehicles is coming to the forefront of public and academic interest. As recently demonstrated by several publicized attacks, an attacker can have a disastrous impact by hacking into vehicles. Traditional defenses such as malware signatures or cryptography face problems in this realm due to resource limitations and real-time requirements of the on-board ECUs.

The current trend of big-data analytics presents a new opportunity for combating this rising security threat. We investigate the use of large datasets to build normal models of the sensors within a vehicle and identify anomalous cases which deviate from the normal model. Furthermore, we develop real-time tools to detect these anomalies during run-time so we can quickly take action to isolate and remedy the source of the anomaly.


Our research pursues two main lines of thought -- (1) vehicles tend to behave similarly in the same location and (2) sensors within the vehicle are connected and thus form a natural redundancy. We approach this problem with techniques from machine learning including Principal Component Analysis for dimensionality reduction and clustering algorithms for unsupervised learning.
Hybrid Vehicular Testbed

Vehicular research is largely driven datasets capturing the aspects of interest to the study. However, once collected, the dataset is frozen in time and any new research questions are limited by the breadth of the data collected. We are building a live testbed to assist in vehicular research. Such platforms have shown to be invaluable in other fields of research such as networking research (PlanetLab) and Mobile use research (PhoneLab). Our testbed is designed with three goals in mind -- (1) multifaceted data collection, (2) flexible interface for researchers, and (3) immediate value for the participants in the testbed.

This project is currently on-going and is expected to be completed in May 2018.
Mobile Sensing
Augmenting Touch Sensitivity with Smartwatch
Signal Processing, 2014-2015
We use sound exchanges between a smartwatch and a smartphone to detect user touch input. This is used to add touch sensitivity to non-sensitive surfaces such as projector screens.

Paper
Hand Gesture Recognition with Microsoft’s Kinect
Machine Learning, course project, 2012
Microsoft's Kinect uses depth and color cameras to identify skeletal structures in pictures. We extended this algorithm to distinguish between different hand gestures.

Course report
Hand Gesture Recognition using Audio Echolocation
Signal Processing, Jan - Apr. 2014
Like dolphins and bats, phones can also play and bounce sounds to learn about their environment. We designed a system to reflect sound from a user's hand to measure the distance and the type of hand gesture held by the user.

Paper
Mobile Security
Mobile WiFi Botnets
Mobile Security, 2012 - 2013
Botnets evolve with evolving technology. Due to mobility, smartphones access the Internet through different access points. This feature makes them appealing as members of a botnet.

I studied the nature of such a threat and proposed countermeasures.

Unpublished paper
Mobile Spynet
Mobile Security, Quals Project, 2013
The pervasiveness of our personal devices (e.g. smartphones) is a blessing and a curse. Each sensor in a smartphone presents a potential privacy leak.

In this work, we investigate the extent of data exfiltration through smartphone cameras and microphones.

Qualifications report