Nine College of Computing & Informatics (CCI) undergraduate students were among the 176 STAR (Students Tackling Advanced Research) Scholars who
presented their research in the annual Summer Showcase held in the Bossone Research Enterprise Building on Aug. 25.
The
STAR Scholars Program allows first-year students to participate in
faculty-mentored research, scholarship, or creative work during the summer after their freshman year. STAR Scholars work full-time during the summer term
while living on campus and earning a $3,500 stipend. The STAR Scholars Program provides an opportunity for students to get to know faculty, explore a major
area of research, and gain practical skills and valuable research experience.
The Showcase is an end-of-summer celebration of the research, scholarship and creative work completed by the Scholars.
The following CCI STAR Scholars participated in the 2016 Summer Showcase:
1) Ryan Howard, BS computer science, and
Anton Hassing, BS computer science
Poster:
Applying TableILP to the Question-Answering Pipeline
Faculty Mentor:
Assistant Research Professor Marcello Balduccini, PhD
Graduate Student, Co-Mentor:
Emily LeBlanc
Natural language processing is a field of computer science that aims to develop algorithms capable of understanding human language. Question answering (QA)
is a natural, language-processing task that uses computer algorithms to answer natural, language questions.
There are several core tasks that a QA system must perform; this sequence of tasks is often referred to as the QA pipeline. Quails is a modular QA
framework under development at Drexel University. Quails breaks down the QA pipeline into individual modules that the user can customize to prototype
end-to-end QA systems.
The goal of my research was to develop and answer selection module for Quails. I studied and adapted an approach recently proposed by researchers from the
University of Illinois and the Allen Institute for Artificial Intelligence. The approach, called TableILP, answers multiple-choice questions by reducing
the task to that of solving an integer linear program.
When fully integrated in Quails, the module will use the natural language question and the candidate answers produced by other modules in the Quails
pipeline to construct an equivalent multiple-choice question that can be processed and solved using techniques derived from the TableILP approach.
My work during the research period involved familiarizing myself with natural language processing and the Quails QA pipeline. I also studied the TableILP
algorithm, implemented it within the pipeline, and conducted experiments to test effectiveness.
2) Klimentina Krstevska, BS computer science/mathematics
Poster:
Automated Visualization and Evaluation of Experimental Plan Recognition Data
Faculty Mentor:
Associate Professor Christopher Geib, PhD
Plan recognition is a research area that fails under Artificial Intelligence and is defined as identifying the actions and goals of one or multiple number
of agents based on a series of observations of the agent’s actions. Dr. Geib and his team of graduate students work on plan recognition system called
Engine for LEXicalized Intent Recognition (ELEXIR). The use of this software results in the output of files containing thousands of explanations. The fact
that each explanation consists of initial and final states (before and after running the software), as well as the likelihood statistics and roots for that
explanation, makes it very difficult for a human to manually differentiate between multiple explanations.
Therefore, I have developed an application that aids the visualization and analysis of data output from ELEXIR. The application displays the most crucial
information about each explanation in separate units, making the data easy to visualize. The application will be able to sort the explanations in order of
their likelihood so that the most relevant ones can be accessed quickly. It will also allow the team to search among explanations for certain patterns and
highlight the differences between two or more explanations. With this application in hand, the team will be able to use the results from ELEXIR in a more
efficient manner.
3) Mosfiqur Rahman, BS computer science
Poster:
Adversarial Stylometry
Faculty Mentor:
Associate Professor Rachel Greenstadt, PhD
Some of the major problems of our academic and corporate environments include plagiarism, copyright infringement, and intellectual property disputes.
Source code authorship attribution has immediate implications for the security community, particularly in its potential to significantly impact
applications like software forensics, plagiarism detection, and determining software ownership. While work on stylometric attribution has achieved great
results attributing system and application programming languages, other types of languages such as scripting languages used for client-side web
developments remain in need of further exploration. In this work, we will present a technique for authorship attribution of source code written in the
common scripting language JavaScript, such as can be acquired through web page source code. We will perform authorship attribution using features derived
from the abstract syntax trees, and will show that the techniques developed for attributing code written in languages such as C++ are generalizable to
other types of programming languages.
4) Guruansh Singh, BS computer science/mathematics
Poster:
Interacting with J-Bob
Faulty Mentor:
Professor Jeremy Johnson, PhD
This project develops a tool and explores strategies to assist students in learning how to derive formal, proofs about software systems. Modern society
relies on computer software to function and the correct operation of software systems is essential. Computer bugs can have costly consequences and are
difficult to detect and fix. Traditionally testing, running programs with different inputs and operating scenarios, is used to find errors in software;
however, since there are infinitely many cases to test, systems cannot be verified through testing.
Alternatively, it is possible to provide mathematical proofs that a software system is correct. The adoption of such strategies has been limited due to the
difficulty in applying formal proofs to real-world software systems, the lack of good proof tools and the fact that students have not been trained to use
such techniques. However, recent progress in the underlying theory and the development of tools called proof assistants had enabled these techniques to be
applied to real-world software and though as part of a standard undergraduate CS curriculum. In this project, we develop an interactive tool, based on the
J-Bob proof assistant, that allows students to more easily derive and explore inductive proofs about properties of recursive functions. The proof tool will
be evaluated in the course Mathematical Foundations of Computer Science (CS 270), taken by CS majors at Drexel in their sophomore year.
5) Safa Aman, BS computer science
Poster:
Assessing Computer Interfaces to Enhance Creative Thinking
Faculty Mentor:
Assistant Professor Erin T. Solovey, PhD
The Drexel Advanced Interaction Research (AIR) Lab has been working to evaluate how computer systems can stimulate creative thinking in users. The AIR Lab
is most interested in the different brain states that appear based on how actively an individual is generating new ideas as he or she performs a creative
brainstorming task. This data is gathered using a lightweight, non-invasive brain imaging technology called functional Near-Infrared Spectroscopy (fNIRS).
The fNIRS device can track oxygen levels in the bloodstream, which then give insight into how the brain behaves given the current state of the subject. The
task is completed on a web interface that allows users to ask for more “inspirations” when they need a new idea. It also logs the timing of user
interactions, such as when the “inspirations” button is clicked or how frequently ideas are submitted. By pairing this rich log data with fNIRS and other
sensors that monitor the heart rate and eye movements of the users while they perform this task, the AIR Lab seeks to paint a complete picture of what
happens during creative thinking.
The ultimate goal of this study is to improve interfaces so that they drive users to generate creative ideas that are richer in both quantity and quality.
While computers and creativity have not always been intuitively associated with one another, the AIR Lab hopes to demonstrate that a finely tuned interface
can prove useful to users who want to think creatively.
6) Enioluwa Segun, BS computer science
Poster:
Live Streaming and Analysis of Biosensor Data to Improve Human-Computer Interaction
Faculty Mentor:
Assistant Professor Erin T. Solovey, PhD
Brain-computer interfaces and biosensors are now widely available, and could provide valuable information to an interactive computer, enabling it to better
support the user’s changing cognitive state. The goal of this project was to live-stream and analyze physiological data from a subject during studies in
the Advanced Interaction Research Lab as he/she performs activities at a computer. The purpose is to determine how the brain reacts to tasking activities
via changes in body variables. These include brain activity, hear rate, skin temperature, galvanic skin response (GSR), and eye movement. Understanding how
the brain works during peak periods would help to create better systems that maximize user performance.
My task was to develop software tools that stream the data as it is collected, converging it into one database. I worked specifically with the Microsoft
Band, which provides sensors for GSR, skin temperature and heart rate. Working with Java, Android SDK, and Band SDK, I built an Android app that directly
taps into the Band data stream and sends the acquired data as Unicode over TCP/IP to a server program running on a PC. The server then uploaded to the
database which ran queries to visualize the data side-by-side across time for easy analysis.
7) Yuvraj Sharma, BS informatics
Poster:
Probabilistic Information Retrieval
Faculty Mentor:
Associate Professor Weimao Ke, PhD
In today’s digital world, we are constantly involved in the seeking, consumption, and generation of information. To satisfy our information needs, we need
effective mechanisms, to help us sort through billions of documents and identify the relevant. Information Retrieval (IR) is the research field focused on
finding relevant information for users. Classic IR models include the Vector Space Model (VSM), probabilistic models based on the Probabilistic Ranking
Principle, and their variations. These models have been shown in prior research to be effective and have been widely adopted in academia and industry.
In this research, we implemented a workflow to parse and index hundreds of thousands of documents from a benchmark. TREC (Text Retrieval Conference) data
set and integrated three classic models, namely TF-IDF, BM25, and Divergence from Randomness, to experiment with their retrieval results. This
implementation is based on a Java search engine library called Apache Lucene. Ultimately, we are creating a variation of BM25 based on a term weighting
scheme derived from the Least Information Theory (LIT) and plan to compare its effectiveness with existing classic models.
8) Merlin Cherian, BS computer science, Velay Fellow
Poster:
Designing a Collaborative System to Improve Family-School Partnership in Special Education
Faculty Mentor:
Assistant Professor Gabriela Marcu, PhD
Family-school partnership has been linked to better student outcomes in special education. Effective partnership includes communication, sharing
perspective, and empowering parents to participate in decision-making. School staff members have traditionally used paper reports to communicate with
parents. However, preparing individualized reports is time-consuming for school staff. On the parents’ part, they find that these reports are not detailed
or reliable enough to help them understand their child’s progress.
The goal of this research is to develop a web application to foster family-school partnership. We are extending Lilypad, a data collection and analytics
tool for school staff, with a parent-side application which will support family-school partnership in three ways: (1) consistent e-delivery of reports
aiding long-term progress monitoring, (2) detailed behavioral reports with easy comprehension and (3) instant messaging for effective communication.
Our user-centered design approach involved a literature review on family-school partnerships, the existing methods of communication and information
exchange and their challenges and barriers. Using inductive thematic analysis, we compared our findings with data from five interviews with parents of
children in special education. We also performed requirements elicitation with two school staff members to understand the school perspective. Our emergent
themes led to an informed ideation of features to address the needs of family-school partnerships. The result of this work is a functional prototype,
Lilypad Home, which has been developed following expert usability evaluation, and will be used in future deployment studies.