Course Home |
Forms |
Project Ideas
|
|
|
|
|
Computer Science 4961/4962
Capstone Project
Fall 2017 |
The first phase of the capstone experience is the selection of a
project and the creation of teams. This page provides a summary of
potential projects for this semester. Students must provide
their personal preferences by
5:00pm Tuesday, September 5, 2017
, so if you wish to suggest
another project idea for consideration, such a suggestion must be
provided by classtime on 3:00pm Friday, September 1, 2017
.
You may wish to take a look at
past project descriptions
that have been selected in recent years.
Table of Contents:
-
A Mobile App for Teaching Pronunciation of
World Languages via Real-time Spectrogram Analysis
-
A Web-based Tool for Drawing Concept Maps in Biology
-
Tablet App for NICU "Code Sheet"
-
Medical Education Project
-
Advanced Data Editor
-
Image Analysis for Paleographic Study
-
Image Realignment for Published Manuscripts
-
Supporting Fréchet Distance in CGAL
-
Financial Risk Management with Apache Spark
-
Very Low Latency Real-Time Computation
-
Power-aware Control on the Raspberry Pi for UAV control
-
Multi-User Virtual Reality Interface for Control of Aerial Drones
-
Collaborative Telepathology
-
Reunifying children with their legal guardians
after (natural or man-made) disaster using serverless computing
-
An Online Gaming App Using Haptics to Teach STEM Concepts
-
Whack a Mole: A Risk Minimization Approach
to Moving Target Defense in the Cloud
-
Evaluation of Machine Translation Approaches for
Closely-related Languages
-
Reboot of Accentuate.us
-
Spellchecker Development
-
Self-contained Search Engine Solution for Irish Language Websites
-
Syntax-aware Machine Translation of English into
VSO Languages
-
SLUber: Technology improvements for the SLU Ride operations
-
Games (tbd)
| Organization: |
SLU's Department of Languages, Literature, & Cultures |
| Client: |
Prof. Christina Garcia <christina.garcia@slu.edu> |
| Supervisor: |
Drs. Kevin Scannell and Flavio Esposito |
| Description: |
Drs. Christina Garcia (SLU) and Terrell Morgan (Ohio State) have built
a set of
web-based tools that take learning far beyond the classroom for L2
pronunciation students. Among these is a site where students record
themselves and can instantaneously compare their own voiceprint to
that of a native speaker. For each sound, the system displays native
and non-native speech samples alongside spectrographic representations
of them. The interface enables students to produce live-generated
spectrograms, re-recording until their produc- tions both sound and
look like the native output. Recordings and student data are
automatically submitted to an instructor interface that allows for
effortless monitoring of students’ progress. This interface was
originally developed for Spanish and has been extended to English and
Quechua, with the eventual goal being applied to many languages.
For this capstone project, students will be required to participate
in the design, implementation and testing of a mobile app and advance
the current web-based platform with a cloud-based system (via Serverless
technology) for rapid and scalable language feedback processing.
| Organization: |
SLU Biology |
| Client: |
Dr. Elena Bray-Speth <elena.brayspeth@slu.edu> |
| Supervisor: |
Dr. David Letscher |
| Description: |
In introductory biology classes, students are asked to create graphical
concept maps
to describe various systems. Currently, the students create these
diagrams on paper, by hand, including starting again when later
revisions are submitted. The goal is to have a web-based system to:
allow students to create, edit, and submit such diagrams
electronically; allow for instructors and teaching assistants to
provide feedback within the system; allow educational researchers to
perform analysis of students' works.
In a previous year's capstone project, CS students created an initial
proof-of-concept, developing both drawing capabilities and a system
with both student and instructor roles. That system was implemented
using a MEAN
stack, yet on a monolithic system. The goal for this new capstone
team will include: to redeploy such a system on a true multi-tier
architecture, with separate web, database, and authentication servers;
to make improvements to the overall usability of the core systems; to
support a test user group and to prepare for a pilot group of actual
biology students.
| Organization: |
Cardinal Glennon Children's Hospital |
| Client: |
Dr. Christopher Brownsworth <christopher.brownsworth@health.slu.edu> |
| Supervisor: |
tbd |
| Description: |
We would like to design a tablet application to function as a "code
sheet." Basically, when a baby is first born or requires CPR (in
medical terms "codes"), one person's job is to record the events and
interventions. Currently, this information is scribbled on the back
of a napkin or scrap piece of paper. Typically, this record is less
than legible and has to be copied into the medical record. We want to
make an application that 1) makes it easy to record the events,
2) easy to export into the medical record 3) simple to analyze
the information. I have sketched out the design and functionality.
There are no similar applications available for this. This could be
expanded from the NICU into adults and pediatric populations.
| Organization: |
Cardinal Glennon Children's Hospital |
| Client: |
Dr. Christopher Brownsworth <christopher.brownsworth@health.slu.edu> |
| Supervisor: |
tbd |
| Description: |
Much of medical education is still conveyed through classroom
didactics and books/paper/ online material. Use of mobile devices has
skyrocketed over the last 10 years; however medical education as a
whole has not embraced mobile devices as a way to teach medicine; We
are currently conducting a randomized controlled trial to determine if
texting medical students daily medical question/answers can foster
learning and improve performance on national standardized testing. We
are limited on the information we can convey through text messaging
due to the character limit. We are wanting to build a platform for
delivering question/answer style questions to students, both online
and mobile devices. This is envisioned as a platform where "classes"
containing question/answer style information can be created and then
students can subscribe to the class and receive the questions either
daily or in block. The system will then create an answer report
breaking down the statistics.
| Organization: |
SLU's Walter J. Ong, S.J. Center for Digital Humanitites |
| Client: |
Donal Hegarty <donal.hegarty@slu.edu> |
| Supervisor: |
tbd |
| Description: |
The IIIF (iiif.io) standard allows discovery and sharing of
datasets. The underlying data format is JSON and the IIIF standard
defines what must or should be a part of the object to make it a
valid IIIF object. The Walter J. Ong, S.J. Center for Digital
Humanities has a data repository for these IIIF objects called
RERUM. RERUM has a web facing interface (rerum.io) through which
various tools interact with these IIIF data objects in the
repository. One of the goals of the site is to offer a robust and
user friendly IIIF Manifest editor through which the rules of JSON
and the IIIF Presentation API
will be enforced while a user creates and/or alters IIIF objects for
the RERUM data store. The RERUM interface is written with HTML, CSS
and AngularJS which are the languages that must be used to complete
this task. A successful implementation of the editor would be
immediately incorporated into the RERUM tool set for public use.
| Organization: |
SLU's Walter J. Ong, S.J. Center for Digital Humanitites |
| Client: |
Donal Hegarty <donal.hegarty@slu.edu> |
| Supervisor: |
tbd |
| Description: |
Beneath the transcription interface at t-pen.org, is a simple image
analysis tool which endeavors to cleanly bound the columns and rows of
handwriting on images of manuscript leaves to enable line by line
transcription by our users. The goal of this project would be both to
improve on the existing algorithm and offer the analysis as a
stand-alone public service. The existing code in written in Java and
may be optimized or replaced. A proposal to port the library into
another language for better performance or more flexibility as a web
service would also fulfill our goals. Support will be offered from the
full-time developers in the Walter J. Ong, S.J. Center for Digital
Humanities who have worked with the original T-PEN project and
code. Sample image sets will be provided; no paleographical knowledge
is required. A successful service will be incorporated into the
existing and next versions of the T-PEN transcription tool and the
public service will be published for use to the academic communities
around manuscript studies and images standards for Digital Humanities.
| Organization: |
SLU's Walter J. Ong, S.J. Center for Digital Humanitites |
| Client: |
Donal Hegarty <donal.hegarty@slu.edu> |
| Supervisor: |
tbd |
| Description: |
Millions of images are made available to scholars through the IIIF
(iiif.io) standard and are organized into virtual containers
(Manifests) by the hosting repositories. For close study of the text,
scholars often need to crop out surrounding support materials (rulers,
color bars, labels), rotate images, split images (of two-page layouts,
for example), and discard uninteresting leaves. Currently, these
adjustments cannot be saved for personal use or publication. The
Walter J. Ong, S.J. Center for Digital Humanities builds tools for
scholars and developers who wish to interact with these materials and
would like to add this one. With the support of the Center’s full-time
development staff, this project balances UI/UX development for simple
user input with implementation of API and standards. Changes users
make will be saved to a public digital object repository (RERUM.io)
and the Manifest which has been augmented will be notified through
Linked Data Notification (inbox-docs.rerum.io). The successful project
will be immediately rolled out to an international community of users
and be made available to the 2000 scholars currently transcribing on
t-pen.org. No experience with manuscripts is required, but comfort
with HTML, CSS, and JavaScript (including AJAX) is.
| Organization: |
SLU CS |
| Supervisor: |
Dr. Erin Chambers |
| Description: |
The Fréchet distance is a fundamental similarity measure on curves in
computational geometry. This project proposes adding Fréchet
distance computations to the Computational Geometry Algorithms Library
(CGAL), which is one of the largest open source geometry toolsets
available. It is expected that the first semester would be spent
understanding an available python implementation of Fréchet algorithms
in 2d and 3d, as well as installing and becoming comfortable with
CGAL, while 2nd semester would be spent in actual code development,
including communicating and working with researchers at the CGAL
project.
| Organization: |
SLU CS |
| Supervisor: |
Dr. Ted Ahn |
| Description: |
Value at Risk (VaR) has been widely adopted in the financial industry
to measure risk. It is used for regulatory compliance, understanding
the risk characteristics of large portfolios, and and making informed
trading decisions. Three common methods of calculating Value at Risk
are variance-covariance, historical simulation, and Monte Carlo
simulation. Monte Carlo simulation can be more accurate than the
simple models, but it requires more computational power. Fortunately,
Apache Spark provides an easy way to scale statistical problems beyond
what a single server can handle. Using Spark and historical stock
data, we will calculate VaR of stocks with Monte Carlo Simulation in
less time.
| Organization: |
SLU CS |
| Supervisor: |
Dr. David Ferry |
| Description: |
Current parallel programming technologies do not provide any guarantee
of the latency with which a computation will execute. Existing
concurrency platforms allow programmers to do parallel programming
without worrying about thread creation or CPU management through
simple directives such as parallel for loops, fork/join semantics,
etc. These platforms were typically designed for large scientific
computations where high-throughput is paramount and specific program
timing and latency is not a concern, so the techniques employed in
their construction do not typically guarantee any worst-case behavior.
Recent applications of parallel computing techniques to real-time
control motivate the creation of a new parallel concurrency platform
with simple and demonstrable timing characteristics. This project will
first characterize the latency cost when existing concurrency
platforms (OpenMP and Cilk Plus) activate threads and distribute work
during execution of a program. Then, we will create one or more
archetype concurrency platforms and measure the latency incurred in
simple and observable systems.
This project will be done in C/C++ on Linux. Prior experience with
parallel programming or program timing is not required.
| Organization: |
SLU CS + Aerospace |
| Client: |
Dr. Srikanth Gururajan <srikanth.gururajan@slu.edu> |
| Supervisor: |
Dr. David Ferry |
| Description: |
Battery power represents a hard constraint on the operational lifetime
of embedded systems, and power is a particular a concern for
compute-intensive applications such as control algorithms that
typically execute tens to thousands of times per second. The
Raspberry Pi provides an attractive computational package for embedded
devices needing to run a full-featured operating system (i.e. Linux),
and is capable of being powered via battery for mobile or remote
applications (e.g. aerial drones). When performance of a system must
be guaranteed it is critical for the software system to be able to
understand the available power resources and use them wisely. This
project, in conjunction with Dr. Srikanth Gururajan of the Department
of Aeronautical Engineering, seeks to characterize the Raspberry Pi
for the purpose of performing control computations in unmanned
autonomous aircraft. In particular it seeks to (1) measure the power
consumption of the Raspberry Pi under typical workloads, (2) develop a
cost model capable of predicting how much energy will be consumed by a
computation before it occurs, and (3) create a software architecture
that allows the system to dynamically trade off power consumption and
system control performance in scenarios where mission objectives or
system constraints may change during runtime. It is anticipated
that this project will be done primarily in C or C++ on Linux. Prior
experience with power monitoring or similar applications is not
required.
| Organization: |
SLU CS + Aerospace |
| Client: |
Dr. Srikanth Gururajan <srikanth.gururajan@slu.edu> |
| Supervisor: |
Dr. David Ferry |
| Description: |
Multiple users interacting with multiple drones inside a Virtual
Reality (VR) environment presents a novel set of challenges in
control, coordination, and system integration. This project builds
upon a single-user prototype VR control system completed as a senior
capstone last year, and aims to extend the functionality to a
multi-user interface that allows two or more operators with VR
headsets and input devices to cooperatively command multiple aerial
drones while maintaining safe flight.
This project, in conjunction with Dr. Srikanth Gururajan of the
Department of Aeronautical Engineering, will involve modifying
existing software to support the notion of multiple users, as well as
dealing with complications that arise such as networking, coordination
between VR interfaces over a (potentially unreliable) network, flight
path deconfliction, collision advoidance, etc. It is also hoped to
implement additional features such as refined flight control software
and long-distance radios that will enable drones in the cage near the
Olive parking garage to be flown from Srikanth's lab in McDonnel
Douglass Hall.
This is a multi-system-integration project that will require work
across Windows, Linux, and various programming languages. Previous
experience with aerial drone hardware or VR is not required.
| Organization: |
SLU Medicine |
| Client: |
Dr. Grant Kolar |
| Supervisor: |
Dr. Flavio Esposito |
| Description: |
Telepathology is the practice of digitizing histological or
macroscopic tissue images based on a glass slide for transmission
along telecommunication pathways for diagnosis, consultation, or
continuing medical education. In the majority of non-trivial pathology
cases, to minimize the response time to the surgeon and the
probability of incorrect assessments, pathologists ask for second
opinions from nearby experts (if available) by physically carrying
glass specimens. Especially to complement underserved geographical
areas, expert pathologists should have remote access to difficult case
assessments via telepathology. Today however, telepathology is
practically unused for the applications that would need it the most:
fast and reliable consultations as well as multi-students live
teaching sessions. Moreover, pathology is nowadays mostly taught via
offline methods or via one-to-one mentor-student specimen
analysis. Remotely recreating the effect of a microscope locally
handled would allow multiple pathologies across federated sites to
collaborate on non-trivial diagnoses. Best-effort Internet connections
are simply not enough to support such applications.
Students in this project will focus on the following aspects of
Internet of Things applied to Telepathology systems:
-
Connect a microscope to the Cloud(Back-end)
using the API and the microscope emulator provided
by an open-source microscope manager software
(micro-manager.org),
create the back-end infrastructure
so that the microscope images can be saved on a cloud server.
-
Front-end Web interface
Using Flask, Django, or any other dynamic web framework,
students will emulate a subset of the existing micro-manager GUI
commands so that input will arrive under the form of web
requests. Connecting the front-end to the back-end, pathologists
will be able to interact, dynamically and remotely, with the
images acquired from (an emulated or a real) expensive microscopes.
-
Demonstration
Adjust the backend system to attach it to a real microscope,
available at the SLU School of Medicine, replacing the emulated
version available in micro-manager.org, to demonstrate remotely
control of images. Input commands include a zoom, pan or focus change.
| Organization: |
SLU Medicine, Cardinal Glennon Children's Hospital |
| Client: |
Dr. Rachel Charney |
| Supervisor: |
Dr. Flavio Esposito |
| Description: |
During a disaster, children may be quickly wrenched from their
families. Research shows that children in such circumstances are often
unable or unwilling to give their names or other identifying
information. Children constitute a vulnerable population and special
considerations are necessary in order to provide proper care for them
during disasters. After disasters such as Hurricane Katrina, the rapid
identification and protection of separated children and their
reunification with legal guardians is necessary to minimize secondary
injuries (i.e. physical and sexual abuse, neglect and abduction). At
Camp Gruber, an Oklahoma shelter for Louisianan’s displaced by
Hurricane Katrina, of the 254 children at the camp, 36 (14.2%) were
separated from their legal guardians. It took 6 months to reunify the
last children, while 70% of the children were with their legal
guardian after 2 weeks. Imagine not knowing for 2 weeks (or 6 months)
if your children are dead or alive! In this project, you will use
serverless computing and other technologies to design and develop
software that will be used by first-responders to minimize children
reunifying time.
The goal of this project is to contribute towards a prototype for a
scalable, federated, public health infrastructure that would exploits
text and image-based analysis to effectively expedite reunification
when children cannot be identified. To this aim, students will design
and implement both a front-end, web- based platform to support a
back-end system for text and image recognition, in which the
computations will be offloaded to a serverless platform. Note that
software for image recognition already exist. The first challenge is
to integrate them with serverless computing platforms for a fast and
scalable platform. Aside from the system integration, the novelty of
this project lays on the design of a novel (children) profile matching
recognition algorithm based on text or images available, as well as in
testing the scalability of such matching operations offloaded to the
serverless cloud platform.
| Organization: |
SLU Mechanical Engineering |
| Client: |
Dr. Jenna Gorlewicz |
| Supervisor: |
Dr. Flavio Esposito |
| Description: |
Hands-on learning experiences have been shown to help students better
understand challenging concepts, particularly those in STEM (Science,
Technology, Engineering, and Math). While laboratories are often one
avenue by which students receive hands-on learning experiences, these
are often discrete opportunities that only occur in a classroom
environment. This project involves building a mobile app (game) that
would allow users to have hands-on learning experiences using their
phone or tablet. The mobile app would interact with the Hapkit (to be
ordered online, and be built), which contains all of the hardware for
building a haptic paddle. (See hapkit.stanford.edu.) The haptic
paddle is a force feedback
joystick that enables users to "feel" what is being displayed on
screen. As the paddle handle moves, digital elements move, and in
turn, send feedback (via a Cloud-based system) back to the user
(players) through the paddle handle, creating a force-feedback
system. One of the exciting challenges lays in the force feedback
being variable based on the medium (e.g. if users are pushing through
water versus pushing on a slippery surface versus pushing through
molasses). The (web or mobile) app will provide an example of a
hands-on learning experience that could be made available to students
at multiple levels, without requiring large infrastructure such as a
university lab.
| Organization: |
SLU CS |
| Supervisor: |
Dr. Flavio Esposito |
| Description: |
Have you ever played the game Whack a Mole? If not, you can try
different versions here.
This project is about
emulating the game between an attacker and tenants of a cloud
provider, renting virtual machines. Cloud providers migrate their
tenants around (moles) to confuse a potential attacker. The more you
stay out, the higher the chances to get "whacked", but the more you
move around, the higher are the overhead costs and disruptions. There
are several types of attacks that can be performed if you are
collocated on the same physical machine of the victim. For example,
with side-channel attacks you could steal sensitive information just
by observing a virtual machine writing behavior. For example, when is
the attacked VM writing on the memory, or by observing power
consumption and electromagnetic leaks. Even a sound can provide an
extra source of information, which can be exploited to break the
system. (See
here.)
Students
are required to design and
implement a moving target defense strategy (the algorithmic behavior
of the mole attempting to minimize the probability of being
"whacked"), migrating real virtual machines with Linux KVM, a virtual
machine management framework. Larger groups may work in competing
sub-teams, emulating the behavior of an attacker trying to learn the
mole strategy (using machine learning techniques).
| Organization: |
SLU CS |
| Supervisor: |
Dr. Kevin Scannell |
| Description: |
This would involve training a number of MT engines,
including phrase-based models like Moses and, say, character-based
models using RNNs (TensorFlow) which could possibly work well for very
closely-related languages. I have data for the Gaelic languages and
for Irish language standardization. Other pairs are possible: Zulu/Xhosa?
| Organization: |
SLU CS |
| Supervisor: |
Dr. Kevin Scannell |
| Description: |
In a capstone project from many years ago, a student developed a web-based
platform (previously hosted at accentuate.us) along with a
corresponding firefox
add-on. Its purpose was to use machine learning to automatically
restore diacritics when entering ASCII-only texts, to make it quicker
and easier to type in more than 100 world languages without extra
keystrokes or a special keyboard. That service is no longer active and
there were scaling challenges since the statistical models needed to
be stored in RAM and this limited the number of languages that could
be supported. With this project, a team will rely on more modern
approaches to scalability to revive and extend the original tools.
| Organization: |
SLU CS |
| Supervisor: |
Dr. Kevin Scannell |
| Description: |
Dr. Scannell has raw data consisting of word lists crawled from the web for
2000+ languages. It would be nice to have a website which could allow
a language community to crowdsource the editing (voting, etc.) of
these wordlists to turn them into spell checkers. Part of this would
be an interactive tool for developing so-called "affix files", without
having to be trained on the technical details, and also a tool for
exporting spellchecking addons for LibreOffice, Firefox,
etc. automatically.
| Organization: |
SLU CS |
| Supervisor: |
Dr. Kevin Scannell |
| Description: |
The idea would be to build this on top of existing open source search
engines like Apache Lucene, or anything derived from Lucene (Elastic
Search, Solr, ...). The issue is to allow indexing according
to eight possible combinations of (standard/non-standard,
mutations/stripped, stemmed/unstemmed), using software developed by
Dr. Scannell, and to package the result in a way that's trivial for
site maintainers to deploy.
| Organization: |
SLU CS |
| Supervisor: |
Dr. Kevin Scannell |
| Description: |
There are various open source packages for doing syntax-aware
statistical MT; maybe the most promising in this context would be an
approach based on tree-to-string transducers, like "travatar".
| Organization: |
SLU Ride |
| Client: |
tbd |
| Supervisor: |
tbd |
| Description: |
The SLU
Ride Program is a primarily student-run operation that provide
safe escorts around campus and the surrounding area. It currently
operates in relatively low-tech way. Technological improvements
might involve providing web-based and mobile apps for
initiating requests for escorts, providing automation to dispatchers
to better optimize the movement of the escorts/vehicles in servicing
calls, and a feedback loop that would provide clients with updates
regarding wait times and locations.
| Organization: |
none |
| Supervisor: |
tbd |
| Description: |
There have been several suggestions from students who wish to develop
computer games as a project, likely within the Unity framework. While
I prefer projects that fill a need for a client, and thus which
involve working with such an end user to define/revise project
requirements, I'm adding this to the preference list so that I can at
least gauge the level of interest and whether to "green light" such a
project.
Michael Goldwasser
Capstone Project, Fall 2017
Last modified: Friday, 01 September 2017
Course Home |
Forms |
Project Ideas