Research Projects

Dynamic Pruning and Quantization to Computational Demand  Details here.

Conducted research with  Dr. Karem at the Univerity of Louisville to create a novel approach for optimizing deep learning models by combining dynamic layer pruning and quantization using Proximal Policy Optimization (PPO). Analyzed neural network layers to dynamically prune less essential components during training. Applied quantization techniques to compress model weights. Integrated PPO to maintain model stability and adaptability throughout optimization. This method helped decrease computational demands allowing models to run on less powerful hardware while still maintaining similar.

Outbreak Detection from Search Data: Development of an algorithm for the early detection of food-borne illnesses using internet search data. Details here.

Developed an algorithm for the early detection of Salmonella outbreaks using search data from Google Trends and Twitter. The project focused on predicting outbreaks earlier than traditional methods by analyzing search trends for symptoms and disease-related terms. Preprocessed data using regression imputation for missing values and geospatial analysis with Folium to map data points across U.S. cities. The model incorporated an Artificial Neural Network (ANN) trained to predict Salmonella cases by correlating search trends with population dynamics data. Enhanced interpretability by also developing a Generalized Additive Model (GAM) to provide stakeholders with understandable insights. Implemented dimensionality reduction using Principal Component Analysis (PCA) to optimize model performance and reduce redundancy. Achieved high accuracy through hyperparameter tuning with random search, leveraging log loss cost function and the BFGS optimization algorithm. This project demonstrated my ability to apply machine learning and data science techniques for public health applications. v 

A low-cost IoT and AI-based integrated system for the detection and treatment of  diseases and deficiencies in crops to maximize the quality and quantity of crops while effectively decreasing pesticide runoff and water usage. Click here for details

Rising global demand for agricultural products necessitates a 119% yield increase by 2050, a challenge compounded by limited arable land and growing populations. This research tackles two main yield-limiting factors: nutrient deficiencies and plant diseases. Using drones equipped with cameras and Convolutional Neural Networks (CNN), this research identifies crop diseases for targeted treatment. IoT-enabled soil sensors gauge nutrient levels, allowing a Neural Network to administer optimal fertilizer amounts.

Development of an algorithm for the early detection of food-borne illnesses using internet search data:  a Generalized Additive Model (GAM) and an Artificial Neural Network (ANN) model to predict outbreaks. Click here for details

Utilizing internet trends, we developed predictive models for foodborne illnesses, which affect 1 in 10 people globally and cost the U.S. $77.7 billion annually. Our Artificial Neural Network (ANN) model predicted 97% of outbreaks four days ahead. Implementing this as the first step in the CDC's detection process could save nearly 27% in economic costs, or about $20.84 billion, while accelerating containment

Title: Secure and Efficient Routing of Wireless Sensor Network Using Blockchain and Deep-Learning-Based Algorithms. Click here for details

Industrial IoT systems, enabled by advances in wireless tech and digital electronics, face security vulnerabilities in their Wireless Sensor Networks (WSNs). In 2021, these vulnerabilities impacted 155.8 million people globally. Our research introduces a blockchain and reinforcement learning-based solution that not only elevates network security but also improves efficiency. Compared to existing systems that fend off 80% of attacks, our approach nearly eliminates all vulnerabilities and is 45% more efficient on average. 

Optimizing Hunger Reduction through Technology and Innovation: Leveraging Machine Learning, IoT, and Mobile App to Streamline Food Bank Supply Chain and Redistribute Food Waste. Click here for details

Imagine waking up every day with a knot in your stomach, not knowing how you will feed yourself and your family. There are 49 Million Americans who are food insecure. Ironically we grow 2.5 times the food we need to feed the whole world. In this research, cutting-edge technology including AI and IoT are used to reduce hunger by streamlining the food bank supply chain and redistributing excess food.

Title: Machine Learning Algorithm (Convolutional Neural Networks) to Detect Icebergs from Satellite Images to Help Reduce Global Warming, Oil Spill and Improve Safety of Arctic Navigation. Click here for details

Melting glaciers and rising sea levels exacerbate climate change, threaten coastal communities, and endanger marine navigation. To address these challenges, we developed a Convolutional Neural Network (CNN) model to detect icebergs in real-time satellite images from Sentinel-1. The model, trained on a dataset of 1,500 elements, achieved 90% accuracy. This technology can provide early warnings for ships and oil rigs, potentially saving billions annually and contributing to climate monitoring.

CAMP: Python package for metagenomic analysis Click here for details

This project is to create a Python package for metagenomic analysis for comprehensive metagenomic studies while ensuring the highest levels of reproducibility and offering an extendable architecture for scalability and adaptability in metagenomics research.

Publications/Conferences