Vir Phoha

Vir Phoha

Bio:

Vir V. Phoha received a Ph.D. degree in computer science from Texas Tech University, Lubbock, in 1992. He is currently a Professor of Electrical Engineering and Computer Science in the College of Engineering and Computer Science at Syracuse University.   Professor Phoha holds fourteen patents and has over 250 publications including six books.  He serves as an associate editor of IEEE Transactions on Social Computational Networks and an associate editor of ACM DTRAP.

Professor Phoha is a Fellow of IEEE; AAAS; NAI; and SDPS. He is also an ACM Distinguished Scientist. He was awarded the 2017 IEEE Region 1 Technical Innovation award for contributions to behavior-based active authentication; outstanding research faculty at Northeastern State University; President’s medallion at Louisiana Tech University among many other honors.

Abstracts:

Health prediction and authentication using gait and hand movement on mobile devices 

Sensor steam data from gait and body movements may be used as a predictor of the onset of diseases.  The way a person walks, moves hands, swipes, or puts pressure on a mobile device while swiping is shown to have unique patterns that can distinguish one individual from another individual and can be used to gauge the health of an individual. This information can be gathered through inertial sensors, such as accelerometer and gyroscope etc. which are present on modern mobile devices. In this talk I will talk about feature extraction from these sensor streams and machine learning algorithms to use these data streams for detection of health-related issues and for authentication.  I will also present some attacks and the corresponding defense(s) on these data streams to manipulate predictors for health and authentication.

Towards a Science of Secure Learning for Autonomous Systems 

Learning or the ability to adapt to changing environments, is fundamental to natural and many man-made systems. The core components of learning in man-made systems are the learning algorithms and the exemplars or the data from which the system learns. I will discuss how learning in man-made systems may be compromised, either through manipulation of the learning algorithms or through manipulation of the exemplars used in learning, and how one may prevent malicious manipulation of learning systems. The ability to adapt to a changing environment offers great advantages but it also poses a serious security threat because one can maliciously manipulate the environment—both training data and algorithms to cause the system to fail or worse adapt to adversaries’ goals. I will present methods and algorithms that show vulnerabilities of man-made systems with the potential of malicious manipulation such as pacemakers that regulate heartbeats,; autonomous drug delivery devices that adjust to the needs of the body; email spam filters that learn words used in spam; web-based recommender systems that learn based on browsing patterns, etc.

Face recognition- Current practices, issues, and technical challenges  

The uniqueness of face makes face recognition ideal biometrics for identification. Typically face recognition involves finding a face in an image and attributing an identity to a face. This talk will cover feature extraction from face images and basic algorithms of face recognition based on geometric features and eigenfaces. This talk will also cover technical issues underlying how face recognition may be spoofed; how artificially generated faces that do not belong to a specific human may be used for various applications, such as privacy, advertising, additional data sets for training; and bias inherent in using face recognition in many social applications.  Finally, the talk will cover technical strategies and challenges to overcome spoofing and bias in face recognition.