CS Night 2019

Department of Computer Science's 18th Annual Computer Science Night 

Thursday, Dec. 5 |  5:30–8:30 p.m.
John Lo Schiavo, S.J. Center for Science and Innovation | USF Campus
This year's research and project abstracts

Graduate Projects

Snaplogic Workflow

Presenters: Shenyi Lu, Pontakorn Pakavaleetorn, Vikas Ramaneti
Sponsor: Professor Greg Benson, SnapLogic
Faculty Advisor: Professor Olga Karpenko, CS690                  

SnapLogic is an Integration Platform as a Service (iPaaS) that enables the creation of application and data integration pipelines using a visual programming language. SnapLogic users can create sophisticated logic without using conventional programming languages or knowledge of cloud server configuration. SnapLogic Workflow extends existing pipeline execution by building a manager that coordinates the execution of these data pipelines based on configured dependencies as a Directed Acyclic Graph (DAG) of tasks. The manager can orchestrate complex data pipelines with custom triggers for downstream dependencies, retry capabilities, and direct the flow of execution of dependent pipelines through user defined logic.

 

Data Modeling Automation

Presenters: Soo Jung Kim, Allison Wong, Manali Patil
Sponsor: Christopher Whittelsey and Michael McLaughlin, PNC Bank
Faculty Advisor: Professor Olga Karpenko, CS690

PNC USF team built a web application that effectively automates and simplifies the process of building different models of financial time series. It brute forces through various modeling forms, performs all the data transformation independently, and generates the top N models for consideration subject to user-defined constraints. The output of selected models would include the statistical characteristics of each model and graphical representation of dynamic backtesting and sensitivity testing.  The application significantly reduces the amount of redundancy and provides controls which reduce the likelihood of human error. Furthermore, it may be used to find candidate models more quickly and accurately, which could be very useful for users in benchmarking models against existing candidate models.

 

Expandi Game Logic

Presenters: Juqiang Li, Shuangjiao Hu, Linjing Shao
Sponsor: Jon Rahoi
Faculty Advisor: CS 690 Master's Project, Prof. Olga Karpenko

Expandi is a web-based version of a popular board game called the Gaia Project, where 14 races from different planets and with special skills strive to colonize the galaxy. To win the game, they need to gain resources, build and upgrade buildings, terraform others’ home planets, form federations, and research new technology. At the end of the game, the player with the most VP wins the game. We implemented the game logic of the game and also built a user interface in Unity3D. We support a multiplayer mode, and store the game history in Firebase.

 

AR-Enhanced Navigation using Mapbox Vision SDK

Presenters: Yanan Liu, Zun Yang, Domingo Huang
Sponsor: Megan Danielson, Mapbox

We developed an iOS app that provides an AR-enhanced navigation and helps people drive more safely by detecting traffic violations such as speeding, running a stop sign, hard braking and acceleration using Mapbox Vision SDK. Our app tracks how well the user drives and calculates the corresponding score. Any detected violation will result in the score deduction, while driving without any violations will lead to a score reward. The app also provides safety alerts while the user is driving, such as a warning about possible collisions. Once the trip is completed, we generate a report that describes traffic violations that occurred and shows where they happened.

 

Improving Adomni Score

Presenters: Kei Fukutani, Tae Lee, Tuo Sun
Sponsor: Chris Weiss, Adomni
Faculty Advisor: Professor Olga Karpenko, CS690

Adomni helps connect advertisers and billboard owners. Advertisers can go on Adomni website and search for the best billboards available to them based on the type of audiences they specify.  Given the billboard data collected by Adomni and the mobile device data from a location intelligence company, we utilized Big Data and Machine Learning techniques to calculate a more accurate Adomni score that is used to recommend the best billboards that have the most exposure to the advertiser’s targeted audiences.

 

Discovering Gravitational Lenses with Deep Learning

Presenters: Yuxing Huang and Ashwini Badgujar
Sponsor: Professor Xiaosheng Huang, Physics Department
Faculty Advisor: Professor Olga Karpenko, CS690

The existing model of finding gravitational lensing has been helping us semi-automate the process of finding new lenses. It’s significant in the field of astrophysics because these systems can help us understand the two biggest mysteries in the universe: dark matter and dark energy. With this mission in mind, in this project, we improved the model code-wise, moved it to production on Google Cloud Platform, and further researched and redesigned the existing algorithm with state-of-art technology to operate on full optimicity.

We've conducted distributed training on Google Cloud Platform using Horovod framework. After running inference on 10 million "REX" type galaxies, we will be able to identify more gravitational lenses than what we currently have. We also looked into transfer learning to optimize the algorithm and make it become more powerful and are moving toward full automation in finding accurate lenses. We also compared the improved neural network models with the existing model to validate if the improvement lives up to expectations. Finally, we packaged the whole project so that researchers and students will be able to easily install it as a library.

 

Sustainability App

Presenters: Nick Kebbas, Bill Li, Shihao Sun
Sponsor: Jose Alvarado
Faculty Advisor: Professor Olga Karpenko, CS690

Currently on the market, there are no popular universal mobile applications for improving one’s sustainability practices. Because of this, we feel there is a need for an application that addresses issues in sustainability, recycling, and environmental friendliness by providing several useful features. By combining features that contribute to our goals of combating food waste, making recycling easier, and providing users with access to eco-friendly alternative products, we hope that our application can help others live a more sustainable life by reducing their carbon footprint. We believe that our product will be a comprehensive sustainability application unavailable in the current market.

 

Who are You?

Presenters: Angela Chen, Ni Luo, Yalei Shi, Yousong Zhang
Sponsor: David Guy Brizan – USF Computer Science
Faculty Advisor: Professor Doug Halperin, CS690

Automatic speech recognition (ASR) systems are trained on the central version of a particular dialect and perform poorly with nonnative speakers. In this project, the team worked on a deep learning system to identify and categorize the nativeness of speakers. The team used machine learning and linguistic approaches such as neural networks, phonotactics, and prosodic analysis to investigate the theory behind why speech sounds, or does not sound, native. The team created a helper website to compare audio samples from corpora and visualize relevant audio features such as vowel formants, Mel Frequency Cepstral Coefficients (MFCCs) images, and spectrograms. In addition, the integrated deep learning system has been built with PyTorch as the backend to an iOS application based on a Flutter app front-end, glued together with Flask. Users of the application can provide a speech sample, and the system will estimate nativeness.

 

ABC News - Election Project

Presenters: Elizaveta Ozerova, Ian Granger
Sponsor: Nick Ross & Daniel Grzenda - USF Data Institute
Faculty Advisor: Professor Doug Halperin, CS690

The goal of this project is to create a platform to give real-time insights and predict election results for ABC News on the night of the 2020 election. To do this, the overall project architecture is composed of three main parts: a data pipeline, a set of prediction models, and a website to display the status of the races. Our team focused on the data pipeline and the website, while the prediction models are being built out by the sponsors at the USF Data Institute. In building the data pipeline, the team created a system that could handle a variety of data types (xml, txt, zip) from dozens of differing sources, while being flexible enough to adapt to unexpected changes on election night. For the frontend website, the team built a robust system to: (1) display the overall status of the election, as well as a detailed breakdown of each race, and (2) monitor the pipeline for any faults. We are excited to show the work we have done in making the election process as transparent as possible, and hope you stop by our poster!

 

Showcase: Automated portfolios for coders

Presenters: Alper Ozdamar, Hiep Bui, Pengfei Song, Tian Rong Liew
Sponsor: Jon Rahoi
Faculty Advisor: Professor Doug Halperin, CS690

Showcase is a social network for the tech community. Showcase provides coders, data scientists, and product designers with a platform to share, explore, and get hired by featuring their top projects in a portfolio. Specifically, Showcase provides project insights and demos, allows exposure to a wider industry network, and offers visibility to employers.

Smart Device Integration Platform

Presenters: Kedar Khetia, Kunal Sonar, Sope Ogundipe
Sponsor: VMware
Faculty Advisor: Professor Doug Halperin, CS690

A proliferation of smart devices has allowed smart home platforms (like Samsung SmartThing, Amazon Echo Dot) to foster. Using mobile/web applications, users can control unique devices or groups of devices specific to a single commercial platform. This project aims to provide a scalable and extensive vendor-agnostic platform to configure custom flows of activity between a variety of devices. End users can choose from configured physical and virtual devices to design workflows of interest. The open-source EdgeX framework is leveraged to handle communication with all physical devices, irrespective of the manufacturer. Our team’s prototype demonstrates the possibility of smart interoperability between smart devices, thus enabling the customer to be free from a single commercial platform ecosystem.

 

IoT Energy Exchange Platform

Presenters: Rozita Teymourzadeh, Anurag Jha
Sponsor: VMware
Faculty Advisor: Professor Doug Halperin, CS690

In the current era, the trend of harnessing renewable resources is on rise; there is a need to better utilize stored electricity. Our team wants to maximize energy utilization by creating a network of devices that can exchange energy efficiently among themselves. With the advent of Internet of Things (IoT) devices and the open-source framework EdgeX, it is now possible for devices to communicate and react to commands without human intervention. Our team has created an Energy Exchange platform that connects a network of energy supply and consumption devices together. The project uses a blockchain to build a trustless and decentralized network of user devices that maintains a history of energy transactions. The implementation results show that the devices can post their energy readings on the EdgeX layer. Those readings are then used to make energy exchange decision among the devices. The proposed transactions are logged in the blockchain. And then devices can supply energy to other devices in the network.

 

Meal Planner

Presenters: Alex Thompson, Dharti Madeka, Drew-Joseph Noma, Jordan Aldujaili
Sponsor: Jose Alvarado
Faculty Advisor: Professor Doug Halperin, CS690

Meal Planner’s goal is to get you into a habit of eating healthy. Meal Planner does this by tracking and managing your meals according to your diet requirements. Meal Planner curates a customized meal plan based on your specified diet, budget and overall health goals. To keep things interesting, Meal Planner will create a new list of meals for you to try every week taking your preferences into account. In addition, you will be able to identify and save your favorite meals whose recipes will then be readily available to you at any time. Each meal will have a recipe with instructions and ingredients available in a shopping list format making an easier supermarket trip for you. Meal Planner is the easiest way to get yourself into healthy eating habits.

 

Undergraduate Projects

Photo Labelling to Improve Recycling

Presenters: Samuel Escapa, Cori Posadas, Martha Salcedo-Suarez
Sponsor: Recology
Faculty Advisor: Professor Doug Halperin, CS490

Contamination in the recycling waste stream reduces the value and utility of the recovered material. Recology has embarked on a machine-learning effort to estimate the percent contamination in order to report how well a customer is sorting their trash. Key inputs for that effort are photos tagged with items of trash, recycling, and compostable material and an estimate of the percent contamination. Using the framework React, our team has developed a Photo Labelling App for use by Recology personnel. This App redesigns the layout from the original app version to improve organization and enhance user interaction, incorporates features to help labelers tag with more efficiency and accuracy, and includes statistics dashboards to display personal productivity and compare productivity between labelers.

 

Behavioral Based Website Generator

Presenters: Ivy An, Simon Lu, Ziling Wang
Sponsor: Inspiration Ventures/Reactful
Faculty Advisor: Professor Doug Halperin, CS490

Reactful is a SaaS application that dynamically changes the users' web experience by injecting Web content based on user behavior (digital body language) while navigating a customized website. In the project, the team designed a new paradigm for creating and generating dynamic websites to leverage Reactful’s capabilities. The team constructed a flexible, generalized, and minimally invasive fragmentation of HTML and templates, and an API-based dynamic web page creator service. The service efficiently assembles the fragments into webpages for desktop and mobile devices deterministically based on navigation, templates, and user behavior. The end objective is for web pages to be dynamically and efficiently assembled on request while supporting a wide range of content layouts.

 

Data Orchestration Custom Task Framework

Presenters: Xue Feng, Jeremy Li, Hao Shen
Sponsor: Mario Lim - OpenPrise
Faculty Advisor: Professor Doug Halperin, CS490

Openprise provides tools that process sales and marketing data. In this project, the team created a generic task that extends the Openprise processing capabilities to support calls to external endpoints to process and cleanse data.  The team implemented a configurable, generic, high performant, fault tolerant, and high capacity library that supports calls to the REST endpoints of Google App Scripting, Google AutoML, AWS Comprehend, and other tools.  In addition, the team is implementing Google App Scripting code to process HTML Table scraping and Google Sheet dynamic data computation.

 

FaceX: IoT Facial Recognition System utilizing EdgeX on Raspberry Pi

Presenters: Christopher Smith, Mushahid Hassan, Olivia Kumar, Tanja Nuendel
Sponsor: VMware
Faculty Advisor: Professor Doug Halperin, CS490

EdgeX Foundry is an open-source platform supported by VMware that provides an infrastructure to build and deploy IoT applications. Utilizing a custom EdgeX instance on a Raspberry Pi, project FaceX provides users with an IoT application that performs facial recognition of pictures taken by a camera module connected to a Raspberry Pi. FaceX analyzes pictures for faces using a convolutional neural network and then pushes its result to a web service that displays information about the matching individual. The main goal is to implement a keyless entry system. This system will be deployed as an open-source software package that anyone can download and deploy to a Raspberry Pi with a camera module to provide a facial-recognition based entry system for a home, shelter, facility, or other restricted-access environment.

 

The Crucible: AI Competitions

Presenters: Alejandro Garcia, Arturo Galvan-Alarcon, Harrison Keeling
Sponsor: Jon Rahoi
Faculty Advisor: Professor Doug Halperin, CS490

Many people want to explore machine learning, but it's hard to find a practical beginner problem to solve.  The Crucible is a training ground for artificial intelligence (AI) solutions to compete head-to-head in a game of "Tag."  The team is gamifying machine learning to encourage friends to compete to create the smartest AIs or strategic models.  Beginners and AI enthusiasts alike will have a great time training their models to navigate the physics of the 2D world to victory! 

 

Parent Portal for SAT Prep Website

Presenters: Rakesh Raju, Rabiga Mukhit, Jimmy Sran, Stephen Strayer
Sponsor: Kato
Faculty Advisor: CS490 Senior Team Project, Doug Halperin

Currently, Kato is developing a curriculum for students preparing for the Scholastic Aptitude Test (SAT), and an interface for students to view their status in the program. However, an interface that parents can use to view progress does not yet exist. The goal of this project is to implement a full web application end to end, allowing parents to view metrics on their student's performance and their overall progression during their time working with Kato.

 

USF Virtual Reality

Presenters: Lucas Alfonzo, Tracy Chen, Enrique Bascur
Sponsor: Insun He & John Bansavich - USF ITT
Faculty Advisor: CS490 Senior Team Project, Doug Halperin

One of the difficulties of being able to visit or tour a campus is the travel time and investment necessary to get there if you happen to live across the country, or even across the world.  The USF Virtual Reality (VR) project allows students and their family members to learn more about the history of the campus from the comfort of their own homes, using the hands-on capabilities of VR. Take a tour with Alan Ziajka, our resident USF historian, learn about USF’s Jesuit values and history, interact with landmarks of USF and take a quiz to confirm your recently acquired knowledge.

 

Convoy: Group Travel made easy with Mapbox SDKs

Presenters: Dhiveshan Chetty, Edmund Wong, Ian Arvizu, Michelle Dong
Sponsor: Mapbox
Faculty Advisor: CS490 Senior Team Project, Doug Halperin

Are you travelling with a group of cars and want to make sure you don’t lose a driver? Convoy solves common group travel issues, such as individual drivers pulling over for pit stops, separating at red lights, or losing other drivers at freeway speeds, by tracking the location of the current leader car using Mapbox’s Navigation SDK. Convoy promotes safety and reduces time for cars in a group to reconnect. Whether you’re driving with the group, or whether you’re remote and want to make sure drivers reach a destination, Convoy offers an immersive Augmented Reality experience. Convoy provides: (1) a 2D map, using Mapbox’s Map SDK; (2) an Augmented Reality (AR) view of the road, using Mapbox’s Vision SDK, including lane detection, car detection; and (3) an AR arrow pointing towards the group leader. With the Convoy app, keeping track of drivers has never been easier!