Maria Chen, Haochen Chen, Yuxuan Zhang, Haowen Weng
The application we plan to make is a 3D Glasses Virtual Try-on application. This program can collect the user's facial data and create a proper 3D-face model with the glasses the user-specified wearing. Users can see how the chosen glass fits on their 3D face. It will help people pick their glasses through the app, which is more convenient than choosing glasses randomly online or walking into the glass shop.
The objective of the project is to provide preliminary results that will contribute to the development of a machine-learning tool that can accurately measure the quality of children’s handwriting based on data from child psychologists. Analyzing the handwriting of children is expensive and time-consuming. Nonetheless, these assessments can lead to early interventions that can dramatically alter a child's development. My project aims to provide the first steps towards a cheaper alternative by providing methods to reduce the amount of children’s handwriting data required for the accurate classification of children’s handwriting using methods such as domain adaptation and transfer learning.
Jake Johnson, Kelsey Peltz
The objective of this project is to train a computer agent to play blackjack with a higher win rate than a player who uses Edward Thorp’s basic strategy. While Edward Thorp's basic strategy is the optimal approach for a player under standard rules, it doesn't consider the effects of individual casino rule variations. Using reinforcement learning algorithms, the agent will incorporate these variations and develop a customized strategy chart that can outperform Thorp's basic strategy under non-standard rules.
Iris Brook, Madeline Nelis, Emily Powers
In June of 2022, the Supreme Court reversed a five-decade-old Roe v. Wade decision, which protected a woman’s right to an abortion. Access to abortion in Louisiana is completely banned with very few exceptions for physical health. We seek to automate an anonymous system, for both the volunteers and recipients, and provide a user-friendly dashboard to alleviate these problems and empower women. We prototype a lockbox where users can pick up emergency contraceptives. The box has an electronic lock that synchronizes with an online dashboard for volunteers to know when they need to replenish supplies in the box. We also build a text bot for users to text for EC, and then receive the location of the lockbox, a lockbox number, and a one-time use combination for the box.
Jonathan Licht, Marisa Long, Anna Schoeny, Ethan Sollender
Court Watch Nola recruits volunteers to observe Orleans Parish Court proceedings to collect data to support their efforts fighting toward a more transparent and ethical justice system. Accessing the city’s court docket data is fundamental in ensuring CWNola fulfills their mission of advocating for justice and accountability in the courts; however, the city’s docket data is extremely disorganized, disconnected, and lacks sufficient search features. Our team worked with CWNola to identify and extract data that can be useful in their work. Using Django, Heroku, and PostgreSQL, our team created a web application using the enhanced data that includes authentication, search input, and search output pages, allowing the CWNola team to download the raw data output, and access data aggregation summaries and visualizations to support the production of CWNola reports and analyses.
Alex Abadi, Hailey Dusablon, Torri Green, Riley Martin
Our project seeks to have an AI generate a murder mystery plot and run it in a video game environment for players to investigate through dynamic NPC dialogue. NPC dialogue is generated by ChatGPT, which allows the player to say whatever they want to NPCs in order to solve the mystery. We are using ChatGPT to generate these narratives and the dialogue within them, and the Camelot sandbox software as a tool with which to present them.
March Madness is a NCAA Division I basketball single elimination tournament consisting of 64 teams, with 63 games played among those teams to determine the winner. Every year, people fill out brackets predicting the winners of each round of the tournament. The March Madness Machine Learning Challenge is a Kaggle competition in which participants compete by creating machine learning models to most accurately predict the outcome of every March Madness game by predicting the probability of each team winning against every possible opponent in the tournament.
The purpose of this project is to automate the scheduling system for the medical residency program of the East Jefferson General Hospital. In the past, the schedules were made by hand and could take weeks to make, and this program would automate this process. This program accomplishes this by taking constraints from the hospital and using AI to generate the best schedule possible.
Aidan Hussain, Raphael Mahari
Our project forecasts Major League Baseball performance to identify players expected to increase their statistical performance significantly. In doing so, we are identifying players who are likely relatively cheaper in free agency (or as trade targets) while still being expected to contribute significantly. We accomplish this by forecasting performance based on primarily non-result-oriented statistics. The system uses regression analysis.
Raphael Deykin, Yali Tiomkin
Our project aims to improve the accuracy of predicting protein-protein interactions using computational tools, which is critical for drug development and scientific research. Experimental structures of proteins can be difficult to derive due to resource constraints such as time and money. Therefore, computational biologists have explored the possibility of computationally predicting protein structures solely from the amino acid sequence. AlphaFold, a highly accurate protein structure prediction algorithm, has made significant progress in addressing this challenge, with the ability to predict protein structures with over 99% accuracy. However, recent studies have shown that AlphaFold-generated protein structures are not significantly better than experimental structures at predicting binding affinities, highlighting the need for alternative approaches. Our study aimed to investigate whether the incorporation of protein flexibility into models would lead to improved accuracy in predicting protein-ligand interactions.
Caroline Casella, Lily Yee
As technology and medicine continue to advance, these two domains have become increasingly intertwined with the ultimate goal of leveraging technological advancement and research to improve patient care in clinical settings. Deep learning and neural networks have great promise within the medical field in regards to earlier disease detection, advancing research for understudied and complex diseases, and biomarker analysis. Advances in these areas could lead to earlier therapeutic intervention and treatment that will improve patient outcomes and give physicians and researchers better tools to fight these diseases. For our capstone project, we aim to improve upon a method named Multi-Omics Graph cOnvolutional NETworks (MOGONET). We will do this by increasing confidence, reducing uncertainty, and adding interpretability with the potential for a final product of a physician/researcher-friendly platform. This work will further the current understanding of the causes of human diseases and important biomarkers.
Julian Esparza, Evan Hendrickson, Rafa Hojda, Max Motz, Carly Presz
Per our service learning requirement, our team researched and identified significant requests and inefficiencies that could be addressed on the government-run site "datadrivennola.gov." Using NLP, Sentiment analysis, and statistical data analysis, we created a data analysis deck identifying specific patterns in user requests to fix problems or add specific features. We incorporated the results of this analysis into a polished and accessible dashboard with in-depth data analysis of some of the most prominent needs expressed by the New Orleans Government. By producing new insights into New Orlean's datasets and presenting them in an easy-to-use format, we have created a useful analytics tool for both "datadrivennola.gov" and its users.
Garrett Gilliom, Mason Boyce, Michael Xu
The PLANIT application helps tourists plan the best route around a city (New Orleans for now) for several chosen destinations within designated time constraints. The application has a modern user interface. The users are able to choose their desired way of transportation, choose from a list of built-in popular destinations based on their yelp reviews, and the calculated best route through our proprietary algorithm that will be presented in apple maps and can be saved offline for convenient usage.
Blake Anderson, Robbie Case, Zack Wellman
This project creates a prediction model that estimates the demand and potential revenue for Uber rides based on historical data. The project uses a Kaggle dataset called "Uber Fares Dataset," which includes various features such as pickup and dropoff locations, pickup times, passenger counts, and fare amounts. The project employs the machine learning library XGBoost Regressor to predict ride demand and fare amounts. The models are evaluated based on R-squared values and Mean Absolute Error (MAE) values, respectively.
Reagan Esteves, Jamie Hartman, Charles Tyndal, and Chenyu Zhao
Our project uses AI to identify skin diseases in people with darker skin tones, addressing racial disparities in current AI recognition, which leads to misdiagnosis. We source our image dataset from sub-Saharan African hospitals. Enhancing images are done through cropping, filtering, and applying boundary boxes and masks with Mask RCNN and Superpixels (Color and Hue). After enhancement, the images are inputed into a AI model for skin disease identification. Our work improves diagnosis accuracy, shortens diagnosis wait times, and speeds up treatment for patients.
Anton De Franco, Evan Dartez
Significant PPP loan fraud was committed during the Covid-19 pandemic. While fraudsters are being caught and prosecuted, we feel as if there is more fraud present than state prosecutions and Department of Justice lawsuits indicate. Using a custom dataset of confirmed fraud cases, we developed a hard voting machine-learning classification algorithm that assisted in detecting fraudulent PPP loan cases. We then expanded upon how to adapt our model to possibly catch fraud in real-time, as well as consider who would benefit the most from our findings.
A Django web application whose main feature is a course recommender system. The system uses item-item collaborative filtering (CF) to produce recommendation scores. Scores are based on user-reported measures of satisfaction, perceived difficulty, grade, and hours studied. Student characteristics such as self-reported aptitude and learning style are also taken into account to facilitate user-user CF.
Kayla Fortson, Laya Kumar, Alexis Preston, Lola Ramineni, Halley Vance
Many college students who temporarily leave their off-campus residences have trouble finding a subletter, and students looking to sublet struggle to find these residences. Our website, University Subletting, offers a platform for students to connect as well as find and provide subleases. The website includes various features, including a homepage with recommended housing, direct messaging, an option to add, edit, and delete listings, and a page of favorited houses.