Faisal N

- Research Program Mentor

Bio

Project ideas

Enhancing Neural Network Performance for Traffic Video Detection Using a Multi-Model Pipeline Approach

This project aims to improve the performance of neural networks for real-time traffic video detection by using a multi-model pipeline approach. The pipeline will break the task into stages, including frame selection, object detection, object tracking, and traffic analysis. By combining specialized models at each stage, such as CNNs for vehicle detection and RNNs for object tracking, the system will reduce computational load, minimize latency, and improve accuracy. Students will experiment with neural architectures and optimize performance for real-world traffic video datasets.

Optimizing NFT Transactions with an Offchain Component for Decentralized Blockchain Applications

This project aims to develop a hybrid onchain-offchain system to optimize the efficiency of NFT (Non-Fungible Token) transactions. While blockchain provides transparency and security, it can suffer from high transaction fees and slow processing times, especially during network congestion. By incorporating an offchain component, this system will offload certain computational tasks and data storage outside the blockchain, reducing onchain interaction and transaction costs. The offchain component will handle tasks such as metadata storage, batch transaction processing, and user authentication. Only critical, final state updates and ownership transfers will be recorded on the blockchain, ensuring both security and efficiency. This architecture will not only decrease gas fees but also improve transaction speeds, making NFT trading more scalable and accessible. Students working on this project will explore blockchain fundamentals, smart contracts, and offchain technologies such as layer 2 solutions or rollups, gaining hands-on experience in decentralized applications.

Modeling Training Performance and Latency in Real-Time Anomaly Detection Using Queueing Theory

This project aims to develop a mathematical model to study the training performance and inference latency of machine learning systems in real-time anomaly detection applications. The focus is on optimizing the efficiency of the machine learning pipeline by modeling the system's performance characteristics using queueing theory. By integrating queueing models, the project will simulate the system's behavior under different conditions, such as varying data arrival rates, model training times, and inference speeds. The model will assess how these factors impact overall system latency and performance, helping to identify bottlenecks and optimize resource allocation. Students will explore how queueing theory principles can be applied to machine learning, especially in time-critical anomaly detection tasks (e.g., cybersecurity or equipment monitoring). The project will provide insights into improving response times and ensuring efficient real-time inference while maintaining high detection accuracy.

Building a Social Network for Environmental Action and Sustainable Living

This project focuses on creating a social network application that empowers users to take collective action on pressing environmental issues and promotes sustainable living. The platform will allow individuals and communities to share ideas, resources, and real-world solutions to reduce their carbon footprint, conserve energy, and promote eco-friendly practices. The application will include features such as: Eco-Challenges: Users can participate in or create challenges (e.g., reducing plastic usage, planting trees) and track their impact over time. Collaborative Solutions: A space for users to discuss and implement projects like community cleanups, local sustainability initiatives, and recycling programs. Environmental Impact Tracking: The app will include a feature that allows users to log their daily sustainable actions (e.g., biking to work, reducing water use) and see their collective impact with others on the network. Students will explore the technical aspects of social media applications, such as user authentication, data sharing, and recommendation algorithms, while also learning about sustainability challenges. The project combines social networking technology with a mission to address real-world environmental problems, motivating users to engage in meaningful, positive change.

Automating Homework and Exam Grading with Large Language Models (LLMs)

This project aims to develop a system that utilizes Large Language Models (LLMs) to automatically grade students' homework and exams. The system will process both structured (e.g., multiple choice, true/false) and unstructured (e.g., essays, short answers) student responses, offering immediate and consistent feedback. For unstructured responses, the LLM will assess factors such as relevance, completeness, and clarity of answers, comparing them to predefined rubrics or model solutions. The system can also offer suggestions for improvement, helping students learn more effectively. Additionally, the project will explore bias reduction, explainability, and the ability to handle diverse subject areas, from math and science to language arts. Students working on this project will gain experience with natural language processing, model fine-tuning, and evaluation metrics, while addressing an impactful real-world need: reducing the grading workload for teachers and providing faster, personalized feedback for students.

Coding skills

Languages I know

Teaching experience

Credentials