The 2018 IUPUI REU Workshop will showcase the research projects conducted during the 2018 NSF/DoD Research Experience for Undergraduates (REU) program at Indiana University-Purdue University Indianapolis. The focus of the workshop is data science and cybersecurity. The workshop will be hosted in ET-202 on August 3rd from 9:00 a.m. to 12:30 p.m.
An Analysis of Secret Image Sharing Techniques and Their Application to Biometric Data Protection
Alexander Z. Eberly1, Khaled M. Graham2
1Computer Information Technology, Indiana University - Purdue University Indianapolis; 2Computer Information Science, Missouri Southern State University
Protecting the privacy of digital biometric data stored in a database has become increasingly more critical in the last decade. The purpose of this study is to analyze the security vulnerabilities of various methods of encoding and decoding user biometric authentication data. Biometric data can include any uniquely identifying biological traits, such as fingerprint, iris, face, gait, or voice verification. This paper summarizes benefits and concerns of several different types of privacy preservation techniques, including visual cryptography and secret image sharing, on facial recognition software. This analysis provides an overview of the most effective techniques for securing user biometric data stored in a database, as well as algorithms for how each method may be implemented.
Mentor: Xukai Zou, Department of Computer and Information Technology, Indiana University - Purdue University Indianapolis;
Analyzing the Dynamic Movement Between Lower Limbs Using Smart Phone Sensors
Kaitlyn A. Hardin1
1Department of Electrical Engineering and Computer Science, School of Engineering and Architecture, Howard University
Motion sensing is a valuable tool to help analyze the movement between objects, and its integration in everyday life is increasingly more apparent. Using motion sensing to analyze health is beneficial and important to health-related advancements. For many individuals, post-surgery healing is necessary for his or her proficient health status. When surgery is conducted on the lower limbs, physical therapy is administered to gauge and evaluate the healing process. However, there is a need for improvement when determining if the healing process is complete. With the aid of motion sensing, sensors available in a standard smartphone can analyze the movement between lower limbs. By accessing the two sensors, accelerometer and gyroscope, we will accurately measure the variances, differentiations, and commonalties between the lower limbs using various features in a machine learning program and effectively determine if the healing process is satisfactory or unsatisfactory.
Mentor: Xiaonan Guo, Department of Computing Engineering and Technology, School of Engineering and Technology, IUPUI
Incorporating Sentence Dependencies into Semantic Word Vector Models for Medical Text Processing
Maia Iyer1, Xiao Luo2
Technology,Indiana University-Purdue University Indianapolis
Natural language processing has been split into two subfields: semantic and syntactic analysis. Semantic processing deals with creating vector representations for each word. The goal is for the vectors to be dense, and that relationships between words is made clear by their vectors. For example, the recently discovered additive property word vectors have (e.g., 'man' + 'royal' = 'king') making analogy problems answerable through these word vector models. The hope for medical language processing is to make more complex relations extractable—for instance causal relations or symptomatic relations. Syntactic processing, on the other hand, deals with gleaning textual information from the structure of the sentence. A currently popular method of syntactic processing is dependency parsing, where connections are explicitly drawn between different words in a phrase or clause, where one word modifies the other. Generally, between these two subfields, there is not much cross-application. The aim of this research paper is to incorporate dependency analysis into a word vector model's training phase to see if this improves the model and its representation of different medical terms. To do so, the Skip-Gram model, a state-of-the-art semantic parser published through word2vec, is modified to grab sentence dependencies and create word pairs based on the dependency pairs. A qualitative word similarity analysis compares the original and modified models.
Mentor: Xiao Luo, Department of CIT, Purdue School of Engineering and Technology, Indiana University-Purdue University Indianapolis
Securing Card not Present (CNP) Transactions
1Department of Computer Science, Purdue School of Science Indiana University-Purdue University Indianapolis
With the implementation of EMV with in-person transactions, the rate of in-person credit card fraud has declined, while the rate of fraud with card-not-present transactions (CNP) has taken off. In order to combat this new dynamic, I have designed a system which utilizes already secure payment systems such as Apple Pay and Google Pay in order to create a way for secure CNP transactions over the Internet. The system uses a module in development from Google which enables bluetooth connections between websites and devices which support Bluetooth Low Energy mode. In this system, a website prompts the user to pair a device. Once the user chooses his device, the website then begins the transaction by transmitting a request for card details from the device’s payment system (Apple Pay, Google Pay, etc) which is returned by if the user verifies the transactions and his identity with a fingerprint. This system contrasts with the current system in which static data is entered into each and every website through which online payments are made. By stealing this information through various methods, a malicious actor can use the credit card information without any further authentication of identity at the point of sale. Implementation of this system allows for the reduction of online credit fraud which greatly decreases the ability of thieves to commit credit card fraud.
Mentor: Xiaonan Guo, Department of Computer and Information Technology, Purdue School of Engineering and Technology,IUPUI
Evading Deep Neural Network and Random Forest Classifiers by Altering Graphlet Frequency Distribution Vectors
Erick E. Bernal Martinez1Bellah Oh2
1Department of Computer Science, School of Science, Indiana University-Purdue University Indianapolis
Machine Learning is a subject which has garnered attention in recent years. Particularly, Deep Neural Networks have been found to be quite useful in tasks such as facial recognition, image classification, self-driving vehicles, etcetera. With this rise in Machine Learning popularity, there has also been a rise in a branch of research known as Adversarial Machine Learning, which focuses on finding potential flaws and solutions encountered with the usage of said algorithms. In this literature, we attempt to attack a Deep Neural Network tasked with malware/benign-ware classification through a series of methods. We compare results attained through poisoning algorithms presented on prior works, while under the assumption the adversary has minimal attacking capabilities. Finally, we explore potential solutions to our attack methodology and draw conclusions from our findings.
Mentors: Feng Li, Purdue School of Engineering and Technology, Indiana University-Purdue University Indianapolis; Xiao Luo, Purdue School of Engineering and Technology, Indiana University-Purdue University Indianapolis
Advanced Sensor-Assisted Facial Recognition Technique to Improve Mobile Authentication
1Department of Computer Science, Westmont College
While facial recognition techniques have become widely used for mobile authentication, common implementations remain vulnerable to attacks such as 2D media attacks and virtual camera attacks. The proposed solution aims to improve security by utilizing the camera to record video to implement nose-angle detection and facial recognition, the accelerometer to compare movements with shakes detected in the video recording, and the proximity sensor to compare distances with accelerometer movements and feature changes in the video. The method is practical, as it does not require any additional hardware for common mobile devices, and it should prove efficient and secure against common attacks on facial recognition.
Mentor: Xiaonan Guo, Department of Computer and Information Technology, Purdue School of Engineering and Technology, IUPUI
Building Concept Hierarchies in the Computer Science Discipline with Representation Learning
Siming Zou1, Luke Schoeberle2
1Mobile Cloud Security REU, Cornell University; 2Mobile Cloud Security REU, Iowa State University
With the national initiative of “Computer Science for All”, it is very important to understand and catalog the concepts and subconcepts of Computer Science discipline in a principled manner. However, there are no existing efforts that achieve this goal. In this research we develop methods for extracting computer science keywords from a large collection of computer science journal article and conference paper abstracts. We then develop methods for identifying relations among these keywords. We are particularly interested in the following relations: isA, part-of, and evolved-to. Using these relations as a guide, we will build representation vectors of all Computer Science keywords, which will subsequently be used for building concepts, and subconcept of computer science discipline at a large-scale.
Mentor: Professor Mohammad Al Hasan, Department of Computer & Information Science, Purdue School of Science, IUPUI