Immersion Experience

Computer Vision “Pedestrian Behavior”

The Computer Vision on Pedestrian Behavior Project aims to analyze and understand the patterns and dynamics of pedestrian movements in various environments using advanced computer vision techniques. By deploying state-of-the-art image and video analysis, this project seeks to identify common behaviors, predict movements, and assess the interactions between pedestrians and their surroundings. The insights gained could be pivotal for enhancing urban planning, improving public safety, and developing autonomous vehicle systems that are better equipped to navigate complex human environments. Moreover, the project's findings have the potential to contribute to the design of smarter, more responsive pedestrian networks in smart cities.

Quantifying Arguments at Scale using Large Language Models

Misinformation can spread because of contradictory evidence in research articles, particularly when predatory venues are considered. In 2020, COVID-19-related research constituted 4% of the world's research papers, making it difficult for researchers and public health officials to remain informed. Moreover, in these articles, there is a substantial amount of contradictory COVID-19-related research published. One article may state an item as incorrect even though hundreds of other articles state it as correct. Hence, to fully understand the research, it is important to understand the scientific consensus of specific arguments (e.g., how many papers make each pro and con argument for a specific claim). For this project, given a specific research topic, we will explore retrieval-augmented large language models to 1) extract claims from research articles, 2) match the claims that state the same argument, and 3) match claims that state the opposite/contradictory argument.

Smart City: Optimized Shaded Routes for Urban Walkers

The Smart City Path Planning with Maximum Shade project leverages digital twin technology to optimize pedestrian routes for maximum shade coverage. This innovative approach utilizes 3D city modeling and solar trajectory analysis to identify and recommend the coolest paths for city dwellers. Aimed at enhancing urban liveability and promoting eco-friendly transport, the project is instrumental for urban developers in strategically planning green spaces and pedestrian-friendly infrastructure. It's a step towards sustainable urban living, providing a dual benefit of comfort for pedestrians and data-driven insights for city enhancement.

Natural Language Processing (NLP) - automated customer support system

Objective: Develop a system that leverages Large Language Models to provide real-time, intelligent customer support across various digital platforms (e.g., web chat, social media, and email). This project aims to create an automated customer support system that utilizes the capabilities of LLMs in NLP to understand customer queries in natural language, process them, and deliver accurate, context-aware responses. By integrating LLMs, the system will be able to handle a broad spectrum of inquiries, ranging from simple FAQ-type questions to more complex troubleshooting and personalized advice. The system will also learn from each interaction, improving its accuracy and efficiency over time. The goal is to enhance customer satisfaction by reducing response times and improving the quality of support, while simultaneously decreasing the workload on human customer service representatives.

Controlling Access to Knowledge in LLMs

As large organizations are fine-tuning company-specific models for their employees, security is becoming more critical. Specifically, not all knowledge should be accessible to all employees. Moreover, it is infeasible to train large models for every sub-group within the organization because limiting the training data could reduce the model’s performance, and tracking thousands of models would be infeasible to scale. Hence, for this project, students will work to implement an access control system within LLMs such that user information (e.g., username, department, etc.) and the organizations current access control rules can be considered before providing a user with a generation. The goal is to develop an end-to-end system that handles access control that is also robust to adversarial attacks.

Machine Learning: Fair House Value Prediction  

The Housing Price Prediction Project aims to develop a machine learning model that can accurately forecast the market value of residential properties based on a range of features such as location, size, number of bedrooms, amenities, and historical sales data. Utilizing various regression techniques and data preprocessing methods, this project will analyze and learn from a comprehensive dataset of housing transactions. The goal is to provide potential buyers and sellers with insightful predictions to make informed decisions in the real estate market. By leveraging advanced algorithms and predictive analytics, the project seeks to enhance transparency, efficiency, and fairness in property valuation, ultimately benefiting both real estate professionals and consumers.

Computer Vision: FaceDeep Detection

The FaceDeep Detection Project aims to develop an advanced facial recognition system that leverages deep learning algorithms to accurately identify and verify individuals in various settings. This project focuses on enhancing security measures, streamlining authentication processes, and improving user experience across digital platforms. By training the model on a diverse dataset of facial images, the system will be capable of detecting subtle facial features and expressions with high precision, even in challenging conditions such as varying lighting, angles, and occlusions. The outcome of this project will be a robust, scalable solution that can be integrated into security systems, personal devices, and access control mechanisms, providing a seamless and secure method of identity verification.

Virtual Reality Safety Enhancement through Egocentric Pose Estimation

Fine Motor Interaction using Full Body Pose Estimation for Children with Autism

Identification of Heat Islands

The Identification of Heat Islands in the City of San Antonio Downtown project seeks to utilize machine learning algorithms and digital twin technology to identify and analyze heat island phenomena within the downtown area. By leveraging datasets encompassing environmental factors, urban infrastructure, and geographical features, this project aims to develop predictive models capable of pinpointing areas prone to heat islands. Through the integration of machine learning techniques and digital twin simulations, the project endeavors to provide insights into the underlying causes of heat islands and propose potential mitigation strategies. The outcomes of this research could inform urban planners, policymakers, and stakeholders in implementing measures to alleviate the impacts of heat islands and enhance the overall livability and sustainability of urban environments.