2022 REU Research Projects

Online Toxic Comment Detection

Faculty Mentors: N. Sertac Artan, Ph.D., Ziqian (Cecilia) Dong, Ph.D., and Huanying (Helen) Gu, Ph.D.

Objective: As social media become ubiquitous, toxicity such as harassment, bullying, and violence are also becoming ubiquitous. However, just because social media is here to stay does not mean that toxicity online has to stay as well. The goal of this project is to develop machine learning methods to improve the identification, and removal of toxic comments from online forums.

Approach: In this project, REU fellows will develop machine learning models of toxic comments online and develop lightweight mobile applications implementing these models to identify and remove toxic comments from online discussions. The fellows will use these models and applications on real-world data and evaluate the feasibility and performance of these approaches.

Mitigating Attacks on Virtual Reality

Faculty Mentors: N. Sertac Artan, Ph.D., Ziqian (Cecilia) Dong, Ph.D., and Huanying (Helen) Gu, Ph.D.

Objective: Virtual Reality (VR) promises to revolutionize many fields such as education, medicine, and gaming by providing immersive experiences. Unfortunately, malicious actors can leverage various aspects of VR to induce physical harm to users. For instance, attackers can disorient users and can cause cybersickness, which presents with symptoms similar to motion sickness and is a result of sensory conflict in the VR environment.

Approach: REU fellows will first evaluate different attack mechanisms for inducing cybersickness, and then develop new approaches to mitigating these attacks. The fellows will assess the efficacy of their approaches with human subject studies.

Design of Novel Reconfigurable Intelligent Surfaces for Secure Wireless Communication

Faculty Mentor: Reza K. Amineh, Ph.D.

Objective: Employing reconfigurable intelligent surfaces has recently gained significant attention in wireless communication. These surfaces are composed of a large number of low-cost units that receive signals from sources, customize them by basic operations, e.g., phase-shifts, and then forward the signal toward desired directions. The use of such surfaces boosts the spectral and energy efficiency of cellular networks without requiring power-hungry and expensive radio frequency chains. All of these properties make reconfigurable intelligent surfaces a promising technology for new standards in wireless communication. The objective of this project is to design reconfigurable intelligent surfaces that can support secure transmission in communication systems.

Approach: In this project, the REU fellow will design reconfigurable intelligent surfaces for secure transmission using high-frequency simulation tools (software) such as Altair FEKO, etc. Through the implementation of this project, the fellow will gain knowledge and experience in the microwave/antenna design process and will learn the relevant critical parameters.

Privacy-Preserving Emotion Recognition

Faculty Mentor: Houwei Cao, Ph.D.

Objective: Design of privacy-aware multimodal emotion recognition systems that preserve emotion-specific information, but eliminates user-dependent information

Approach: Students will first investigate the interplay between emotion-specific and user identity-specific information in audio-visual emotion recognition systems, and further develop a privacy-aware emotion recognition system that preserves high emotion recognition performance without leaking of the user identity. They will gain hands on research experience by designing, implementing, and evaluating the proposed privacy-aware multimodal emotion recognition system.

Smartphone User Authentication with Swipe Keyboard Gestures

Faculty Mentors: Kiran Balagani, Ph.D., and Paolo Gasti, Ph.D.

Objective: Evaluate the security and Resilience of Biometric Traits Extracted from Swipe Keyboard Gestures for Active Authentication of Smartphone Users.

Approach: We aim to develop algorithms to authenticate smartphone users using swiping gestures made on a virtual keyboard. We will extract biometric traits based on the spatial (e.g., gesture shape and position), kinematic (e.g., gesture speed and smartphone movement), temporal (e.g., frequency and duration of gestures), and perceptual (e.g., the time between a gesture and a word appearance) characteristics of the gestures. We will analyze the authentication performance of the traits using a new dataset containing gesture patterns and stress-test these traits against state-of-the-art forgery attacks. By systematically evaluating the new gesture-based biometric traits for active authentication, we will create new knowledge that can potentially reshape smartphone users’ security.

Energy-Aware Privacy-Preserving Protocols on Smartphones

Faculty Mentors: Kiran Balagani, Ph.D., and Paolo Gasti, Ph.D.

Objective: Investigate novel password-based encryption techniques suitable for mobile devices, that leverage deep neural networks to provide security against password guessing attacks.

Approach: We propose to develop a new password-based encryption tool for images that combines generative deep neural networks with password based symmetric encryption to provide security beyond the brute force bound. With modern password-based symmetric encryption schemes, the effort needed by the adversary to decrypt a password-protected image is proportional to the complexity of the password chosen by the user. As a result, the security of these scheme when used with low-entropy passwords is very limited. In contrast, in this project we will develop a new set of encryption tools that do not suffer from this limitation.

Mobile Malware Detection in Smartphones

Faculty Mentor: Wenjia Li, Ph.D.

Objective:In this project, student(s) will get involved in the research efforts on how we can identify malicious mobile applications (malware) in an effective and efficient manner. Students will have the opportunity to design and develop new approaches that first collect and analyze various features related to the Android system, such as the permission requested, network traffic usage, battery usage, memory usage, etc. Based on those attributes, some machine learning or data mining algorithms will be applied to automatically distinguish benign (normal) mobile applications from malicious (unwanted) mobile applications.

Approach: Students are expected to first collect and record the attributes of the Android phones using mobile applications, and then use machine learning/data mining techniques to analyze these attributes and build a classification model to distinguish malicious applications (malware) from normal ones.

Fast and Secure Non-Intrusive mm-Wave Scanning Mechanisms

Faculty Mentors: Anand Santhanakrishnan, Ph.D. and Reza K. Amineh, Ph.D.

Objective: Microwave (300 MHz-30 GHz) and millimeter wave (30 GHz-300 GHz) imaging (MMI) technologies can penetrate inside optically opaque media for inspection and imaging purposes, for applications such as medical sensing and imaging, radar imaging, non-destructive testing, etc. In some of these applications (such as tracking objects behind a wall) the imaging time need to be as short as possible to allow for real-time or quasi real-time monitoring of the inspected domain.

Holographic imaging is a popular MMI technology used extensively in security scanning of passengers at airports. Mechanical scanning of the an antenna in a raster manner over a large aperture is time-consuming and not suitable for fast imaging in applications such as rescue operations, object tracking behind a wall, etc. This warrants development of an efficient algorithm for enabling faster scanning of a raster, which is the main objective of the proposed research project. The key shortcoming in existing systems is that the scanning is performed according to a pre-determined trajectory, which could result in large latency if the object is not located in the initial part of the scanning process.

Approach: This project has the following key outcomes: (1) Identifying linear redundancies (correlations) in the received signal strength of the mm wave electromagnetic signals at different physical distances from the actual location of the object; (2) Utilizing the identified correlations to develop vector quantization and Markovian models to adaptively vary the speed of scanning at different spatial locations (3) Exploiting the identified vector quantization and Markovian models in conjunction with depth and breadth first search algorithms to develop optimal (or sub-optimal heuristics based) trajectories for scanning; (4) Development of a simulation tool to demonstrate the proposed algorithms;

Human Factors in Security

Faculty Mentor: N. Sertac Artan, Ph.D.

Objective: Many failures of online security methods are due to human factors. For instance, online criminals can gain unauthorized access to computing resources using social engineering techniques. Our aim in this project is to evaluate user behavior in security-critical settings with objective methods, namely, by sensing their brain activity.

Approach: The REU fellows will learn to use neurophysiological and neuroimaging techniques such as Electroencephalograpy (EEG) or Functional Near-Infrared Spectroscopy (FNIRS) to monitor brain activity of subject behavior in security-critical settings. The fellows will then use machine learning methods to evaluate the brain activity of the subjects with how the subjects act while making decisions during security related tasks.

Smartphone Geolocation

Faculty Mentor: Ziqian (Cecilia) Dong, Ph.D.

Objective: Investigate effective smartphone geolocation and data analysis techniques to address the geolocation challenges in metropolitan areas when GPS does not function well due to lack of line of sight and interference.

Approach: Students will participate in network measurements data collection such as network delay, bandwidth, data transmission rate, received signal strength and geographical location using applications on smartphones and apply machine learning and optimization algorithms to analyze and predict location of smartphones.

Privacy-Protected Automated Medical Data Collection on Smartphones

Faculty Mentor: Huanying (Helen) Gu, Ph.D.

Objective: Raise the awareness of students to privacy and anonymity issues related to the collection and manipulation of medical data and design a framework that manipulates health-related information and provides strong privacy guarantees.

Approach: Students will develop a tool to automatically extract and de-identify collected medical data. Furthermore, they will identify the critical privacy problems in the existing system, and architect a more privacy-friendly system for automated medical data collection.