In the Summer of 2024, CITIES will activate several PPTP (Post-Graduation Practical Training Program) positions to explore new areas of investigation and promote collaboration with other NYUAD research centers and labs.
Modification of Online News Articles to Instagram Posts
Student: Alex Chae
Supervisor: Kinga Reka Makovi
Description: This project examines how online news articles are modified into an Instagram post and how this impacts audience perception of news. The dataset collected includes approximately 30,000 Instagram post images and captions from six prominent news accounts and the original news articles from which the posts were adapted. Through textual analysis methods, we have found that when online news articles are modified as an Instagram post, the emotional intensity of the content neutralizes, and the frames used to report a piece of news are often modified. The next steps focus on examining whether such modifications make a difference in how the audience reads news. This will be conducted via a randomized survey experiment in which respondents are assigned to either a treatment group for reading an online news article or a treatment group for reading an Instagram post. Both groups will be asked comprehension questions and deeper questions about their feelings and opinions towards the reported news, and these answers will be compared to draw conclusions. This project advances our understanding of the kinds of news that an increasing number of people are reading on social media and how cultural production, such as a news story, is adapted to different platforms.
VM²FL4H: Vertical Multimodal Federated Learning for Healthcare
Student: Omar El Herraoui
Supervisors: Farah Shamout & Michail Maniatakos
Description: The project is aimed at using AI to better predict diseases like cancer by looking at different types of medical data, including X-ray images and electronic health records, from many hospitals. The core of the research revolves around addressing the limitations of current AI models, which predominantly rely on data from single institutions. The goal is to develop a collaborative learning framework, termed Vertical Multi-Modal Federated Learning For Health (VM²FL4H). The framework, VM²FL4H, will allow AI models to use a variety of data from different hospitals (which we call “multimodal” data) to make better predictions about patient health. The team is doing this in a way that doesn’t risk patient privacy. VM²FL4H allows hospitals to collaborate and improve AI models without actually having to send any of their sensitive raw patient data out of their systems. This means the team can learn from a huge pool of data without compromising on privacy. In collaboration with esteemed partners such as Mubadala’s integrated healthcare assets and the Cleveland Clinic Abu Dhabi (CCAD), VM²FL4H is designed to meet the requirements of real-world healthcare settings. The role of a student researcher encompasses several responsibilities, including the integration of privacy-preserving measures into the VM²FL4H framework. This entails a thorough evaluation of privacy-preserving approaches, balancing privacy, accuracy, and complexity/scalability and contributing to the writing and publication of findings. Furthermore, the student will partake in the release of an open-source implementation of VM²FL4H, ensuring that the contributions are accessible to the wider research and healthcare communities.
Efficient and Robust DNNs for Mobile Advanced Driver Assistance Systems
Student: Zaeem Shahzad
Supervisor: Muhammad Shafique
Smartphones provide a unique opportunity to make Advanced Driver Assistance accessible to everyone through an application powered by deep learning models. However, the key challenge in developing such an application is to create low-cost models for all the functionalities involved in an Advanced Driver Assistance System (ADAS). This project will explore methods for building low-cost application-specific DNNs for distance estimation, lane detection, weather monitoring, and driver monitoring. Based on the performance evaluation, the student will investigate different improvement strategies. Finally, security vulnerabilities inherent in the developed vision-based models, particularly in studying the potential for exploitation through backdoor attacks, will be addressed. Since these vulnerabilities pose significant risks to safety-critical systems like autonomous vehicles, it is imperative to employ existing defense strategies to ensure the reliability and integrity of the developed models. In summary, the networks will be optimized primarily for efficiency, with an additional focus on robustness.
CityGraph Tools
Student: Abdul Samad Gomda
Supervisor: Djellel Diffalah
The CityGraph Tools project consolidates and enhances the CityGraph suite, focusing on the development of Kosmograph, an innovative knowledge graph visualization tool tailored for large-scale data sets.
Urbanism and the Anthropocene: An Exploration of Abu Dhabi’s Public Spaces
Student: Aparna Rajeev
Supervisor: Laure Assaf
In this research project, the student will be working with the Abu Dhabi Public Spaces Project, a part of the Anthropocene: Urbanism, Environment and Sustainability Research Kitchen. The project evaluates the transformation of urban public spaces of Abu Dhabi in the era of the Anthropocene and examines the tensions between the design of public spaces and their lived uses/accessibility. Seeking information on the myriad ways through which public spaces are conceptualized, used, and even appropriated by the residents of Abu Dhabi, the project delves deeper into how environmental and social factors such as nationality, race, class, and gender shape access to these spaces. Employing primarily ethnographic research, the project also involves incorporating modes of practices such as mapping, photography, soundscapes, etc., to analyze how placemaking by the residents in these public spaces, both indoors and outdoors, is altered and exhibited in times of climate emergency.