Summer AI Research Institute

A Flagship CEAMLS Program

Nine student research teams present their summer research at NSEA 2026.
Final Presentations: Sunday, July 26, 2026 | Room 210A-B, University Student Center

Explore the Projects

About the Institute

Program Description

The Summer AI Research Institute, held at Morgan State University, is designed to introduce undergraduate students to theoretical and applied research in socially responsible and trustworthy AI, with a focus on addressing complex real-world challenges. The primary objective of the Institute is to enhance the diversity of thought in AI by actively recruiting and integrating interdisciplinary talent from underrepresented groups.

Our overarching aim is to deliver a transformative summer program that offers a comprehensive, in-person AI research experience for a diverse cohort of undergraduate students. This program, tailored to accommodate varying levels of experience, leads participants through the complete research journey, from project selection to the delivery of oral presentations. Through this immersive experience, we aim to cultivate a rich, inclusive environment that fosters collaboration, innovation, and the development of ethical, responsible AI solutions.

Research Projects

Student Researchers

Faculty Mentors

Presentation Sessions

Final Presentations

All presentations take place on Sunday, July 26, 2026 in Room 210A-B at the University Student Center, across three sessions. Schedule is subject to change.

Session 1

Morning Presentations

10:25 AM to 12:00 PM

Project 3 Session 1 · 10:25 AM to 12:00 PM · Room 210A-B

Post-Disaster Building Damage Assessment AI

90-Second Project Spotlight

Mentors

Faculty MentorDr. Steve EfeCivil & Environmental Engineering

Graduate MentorJohn TanimolaCivil Engineering

Project Summary

This project presents an AI-powered framework for rapid post-disaster building damage assessment using the xBD (xView2) benchmark dataset. High-resolution pre- and post-disaster satellite image pairs are analyzed to classify structural damage into four standardized categories: No Damage, Minor, Major, and Destroyed, which are subsequently mapped to ATC-20 rapid evaluation levels (Green, Yellow, and Red) to support emergency response and decision-making. The framework implements a complete deep learning pipeline, including data preprocessing, data augmentation, class imbalance mitigation, and transfer learning. Three deep learning architectures, a ResNet50 baseline classifier, a Siamese ResNet50 network, and a Vision Transformer (ViT-B/16) were trained and systematically evaluated using Overall Accuracy and weighted F1-score. The resulting models demonstrate the potential of automated damage assessment systems to support rapid building triage and improve post-disaster situational awareness.

Student Presenters

Alameen AdekuComputer ScienceSoutheastern Louisiana University
Christian MessadoComputer ScienceMorgan State University
Keren GilHospitality ManagementAnne Arundel Community College
Asheley MudzingwaComputer ScienceOhio Dominican University

Project 1 Session 1 · 10:25 AM to 12:00 PM · Room 210A-B

Cross-View Multi-Graph Contrastive Learning with LLM-Guided Knowledge Initialization for Predicting Antidepressant Treatment Response in Major Depressive Disorder

90-Second Project Spotlight

Mentors

Faculty MentorDr. Jamell DaconComputer Science

Graduate MentorRicky GoleAdvanced Computing

Project Summary

This project addresses a critical challenge in multi-omic integration for predicting clinical antidepressant response. Conventional multi-modal prediction models assume that all molecular data modalities contribute meaningful, task-relevant information. However, this work demonstrates that introducing an uninformative, high-dimensional modality can lead to modality dominance, a failure mode in which the integrated model performs substantially worse than the strongest individual modality. This degradation was observed consistently across multiple data fusion approaches. To mitigate this issue, the researchers propose Signal-Gated Omic Fusion (SGOF), a framework that estimates each modality's predictive utility using held-out validation data and dynamically assigns weights before integration. By suppressing uninformative modalities while emphasizing informative ones, SGOF restores predictive performance and improves the robustness of multi-omic integration without requiring prior knowledge of modality relevance.

Student Presenters

Sharon IwehaHospitality ManagementMorgan State University
Oluwasegun Soji-JohnComputer ScienceMorgan State University
Gabriel ChambersComputer ScienceAlabama A&M University
Peyton BlakeBusiness ManagementDelaware State University

Project 4 Session 1 · 10:25 AM to 12:00 PM · Room 210A-B

Artificial Intelligence for Early Prediction of Cardiovascular Disease Using ECG Signal

90-Second Project Spotlight

Mentors

Faculty MentorDr. Timothy OladunniComputer Science

Graduate MentorFarouk Ganiyu-AdewumiAdvanced Computing

Project Summary

This project proposes an AI-based framework for early cardiovascular disease (CVD) risk prediction by integrating electrocardiogram (ECG), photoplethysmography (PPG), and respiratory signals. CVD remains one of the leading causes of death worldwide, with many conditions progressing silently before clinical symptoms appear. ECG captures the heart's electrical activity, PPG provides complementary information on vascular function and blood flow, and respiratory signals offer additional physiological context to improve prediction when the primary modalities disagree. Machine learning and deep learning models will be developed to extract multimodal features, classify cardiovascular risk, and evaluate predictive performance and stability using the Prediction Error and Confidence Stability (PECS) framework. Publicly available datasets will be used for training and validation, with the goal of developing an accurate, interpretable, and robust decision-support system for earlier cardiovascular risk detection and preventive care.

Student Presenters

Kosisochukwu OgbuanyaComputer ScienceFisk University
Sanaa ReevesElectrical EngineeringMorgan State University
Sandy AkoyManagement Information SystemsUniversity of Houston
Rochak GhimireComputer ScienceMorgan State University

Session 2

Early Afternoon Presentations

1:00 PM to 2:45 PM

Project 2 Session 2 · 1:00 PM to 2:45 PM · Room 210A-B

Developing Edge AI Applications for Human-Robot Collaboration in LEGO Assembling Operations

90-Second Project Spotlight

Mentors

Faculty MentorDr. Yuhan JiangCivil & Environmental Engineering

Graduate MentorDerrick MirindiArchitecture & Planning

Project Summary

This project develops edge AI applications that enable a robotic arm to autonomously identify, retrieve, and manipulate LEGO bricks from a mixed pile to support automated assembly tasks. The research integrates computer vision, image processing, edge AI, and robotics to develop object detection and manipulation algorithms that operate efficiently on resource constrained hardware. Machine learning models will be implemented and evaluated on Linux based edge computing platforms while integrating sensors and robotic control systems for real-time perception and manipulation. The project aims to develop robust methods for LEGO brick classification, sorting, and autonomous retrieval, while evaluating system performance under realistic operating conditions. The resulting system will advance edge AI enabled robotic manipulation and provide a platform for intelligent automation and hands-on robotics research.

Student Presenters

Toluwani OlasokoElectrical EngineeringHoward University
Joaquin SnowdenMechatronics EngineeringMorgan State University
Kayla HicksCybersecurityMorgan State University
Abdulahi OyebanjiComputer ScienceUniversity of Southern Mississippi

Project 6 Session 2 · 1:00 PM to 2:45 PM · Room 210A-B

DermaBridge: A Trustworthy Multimodal AI App for Skin Lesion Screening, Explainable Risk Assessment, and Smart Clinical Referral

90-Second Project Spotlight

Mentors

Faculty MentorDr. Saroj PramanikBiology

Graduate MentorBlessing AdeikaComputer Engineering

Project Summary

This project develops DermaBridgeAI, a fairness first AI framework for skin lesion classification that prioritizes equitable diagnostic performance across Fitzpatrick skin types (FST I-VI) alongside overall predictive accuracy. Existing AI models for skin cancer diagnosis often underperform on darker skin tones, contributing to delayed diagnosis and poorer melanoma outcomes. The project integrates six public dermatology datasets into a unified benchmark and systematically evaluates convolutional neural networks and Vision Transformer architectures using a standardized training and evaluation pipeline. Model performance is assessed using both traditional classification metrics and fairness measures to quantify disparities across skin tones. The resulting framework identifies transformer-based models as promising candidates for reducing performance gaps while maintaining high diagnostic accuracy, advancing the development of more equitable AI systems for dermatological diagnosis.

Student Presenters

Leona FrancisComputer ScienceMorgan State University
Kosisochukwu ObioraComputer ScienceRust College
Bishop AkalusiComputer ScienceMorgan State University
Ayomide AisidaComputer ScienceBowie State University

Project 8 Session 2 · 1:00 PM to 2:45 PM · Room 210A-B

Optimization Driven Machine Learning Framework for Predicting Respiratory Health Risks Using Urban Environmental Data: Baltimore as a Case Study

90-Second Project Spotlight

Mentors

Faculty MentorDr. Olaniyi IyiolaMathematics

Graduate MentorAmara EzeMathematics

Project Summary

This project investigates how optimization driven machine learning can be used to predict respiratory disease risk from environmental factors in urban settings. Respiratory diseases such as asthma, bronchitis, and chronic obstructive pulmonary disease (COPD) affect millions of people worldwide, and environmental conditions including temperature, humidity, traffic activity, and population density can substantially influence respiratory health. Using Baltimore, Maryland, as a case study, the project analyzes environmental datasets to identify patterns associated with respiratory disease risk. Machine learning models, including Extreme Learning Machines (ELM), Support Vector Machines (SVM), and Random Forest, will be developed and compared, with the Douglas optimization algorithm applied to improve SVM training efficiency and predictive performance. The project aims to evaluate the effectiveness of optimization enhanced machine learning for respiratory risk prediction while advancing data driven approaches to public health monitoring and early disease detection.

Student Presenters

Jaren AllenCivil EngineeringMorgan State University
Chanel GreenFamily & Consumer ScienceMorgan State University
Maya MukabeInformation SystemsGrambling State University
Christopher DanielsElectrical EngineeringMorgan State University

Project 7 Session 2 · 1:00 PM to 2:45 PM · Room 210A-B

Predictive Stability vs. Fairness Instability in Clinical ECG Classification: A Multi-Run Analysis of Deep Learning Models

90-Second Project Spotlight

Mentors

Faculty MentorDr. Blessing OjemeComputer Science

Graduate MentorSudip SharmaAdvanced Computing

Project Summary

This project investigates the fairness and reliability of deep learning models for automated electrocardiogram (ECG) classification using the PTB-XL dataset. Four deep learning architectures are developed and compared for multi-label classification of five cardiac conditions. Model performance is evaluated using AUROC and F1-score, while fairness is assessed across intersectional age and sex groups using metrics such as the False Negative Rate (FNR). The project also examines the consistency of fairness across multiple training runs, evaluates class weighting as a fairness mitigation strategy, and assesses model robustness under different types of ECG noise. Grad-CAM is used to improve model interpretability by highlighting the ECG regions that influence model predictions. The findings aim to support the development of fair, reliable, and transparent AI systems for clinical decision support and healthcare applications.

Student Presenters

Stephanie EgwuchukwuPhilosophyMorgan State University
Bto BhattaNursingMorgan State University
Kayvon ShearedPsychologyXavier University of Louisiana
Kenzi MerchantElectrical EngineeringMorgan State University

Session 3

Late Afternoon Presentations

3:00 PM to 3:50 PM

Project 9 Session 3 · 3:00 PM to 3:50 PM · Room 210A-B

AI-Driven PM2.5 Prediction Using Satellite and Meteorological Data

90-Second Project Spotlight

Mentors

Faculty MentorDr. Xiaowen LiClimate Science

Graduate MentorRokeya SiddiquaAdvanced Computing

Project Summary

This project develops machine learning and deep learning models to predict surface particulate matter (PM2.5) concentrations in the Baltimore/Washington, D.C. metropolitan region by integrating satellite-derived Aerosol Optical Depth (AOD) measurements with ground-based meteorological observations. Environmental data collected from the Howard, Padonia, and Beltsville monitoring stations between 2019 and 2022 are synchronized, cleaned, and merged to create both hourly and 30-minute datasets. A three phase modeling framework is implemented, beginning with baseline models, followed by temporal feature engineering using lag and rolling-window features, and concluding with mutual information-based feature selection to identify the most informative predictors. Nine machine learning algorithms and five recurrent deep learning architectures, including Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), and bidirectional variants, are evaluated using RMSE, MAE, and R². The project aims to determine how temporal feature engineering, feature selection, and data resolution influence predictive performance, ultimately advancing data-driven approaches for air quality monitoring and PM2.5 forecasting.

Student Presenters

Blessed KutyauripoElectrical EngineeringMississippi Valley State University
Julien NgandoPhysicsMorgan State University
Eddie King-HedgspethElectrical EngineeringMorgan State University

Project 5 Session 3 · 3:00 PM to 3:50 PM · Room 210A-B

AI-Driven Discovery of Drug Candidate Compounds to Slow Cognitive Decline in Alzheimer's Disease

90-Second Project Spotlight

Mentors

Faculty MentorDr. Roshan PaudelComputer Science

Graduate MentorRamisa FarhaAdvanced Computing

Project Summary

This project develops an AI-driven computational pipeline to identify and prioritize candidate compounds for Alzheimer's disease. The team focuses on five key therapeutic and drug-delivery endpoints: BACE1, AChE, ASK1, Tau, and blood-brain barrier permeability. Public molecular activity datasets were cleaned, standardized, and used to train classical QSAR and ChemBERTa-based models for activity, potency, and BBB prediction. Model reliability was evaluated using scaffold-based internal validation, Y-randomization, applicability-domain analysis, and training-overlap filtering. Candidate compounds were then ranked through an integrated scoring system combining classical QSAR predictions, ChemBERTa molecular representations, CNS-relevant properties, and ADMETlab 3.0 toxicity screening. The final goal is to generate a prioritized, safety-aware shortlist of compounds for molecular docking and further experimental investigation, while using AI responsibly as a screening and prioritization tool rather than as direct clinical evidence.

Student Presenters

Austin HinsonComputer ScienceColby College
Jonathan CoxIndustrial EngineeringNorth Carolina A&T State University
Ranjish KumarComputer ScienceMorgan State University
Destiny BertierComputer ScienceMorgan State University

SAIRI, Poster, and K-12 awards will be presented in Ballroom A & B at 4:00 PM on Sunday, July 26. Schedule is subject to change. See the full Day 2 schedule.

For questions about SAIRI, contact: nsea@morgan.edu