University of Rwanda and CMU-Africa Collaboration Research Seminar

October 08, 2025

12:30 p.m. CAT

A203

TSANet, a unified and robust AI image detection framework

Speaker: Gabrial Zencha Ashungafac, Research Associate (CMU-Africa)

Abstract: Face-based authentication technologies are increasingly adopted in identity verification systems due to their convenience and ease of use. However, they remain vulnerable to a wide range of spoofing attacks. To address this, we propose TSANet, a unified and robust detection framework that combines contrastive feature learning with a probabilistic decision model. Unlike many existing methods that depend on video inputs or train separate models for different attack types, TSANet operates solely on still images and generalizes across diverse spoofing scenarios using a single model streamlining deployment and enhancing scalability. Facial inputs are projected into a semantically structured embedding space using a margin-based triplet loss, which encourages clear separation between genuine and spoofed samples. To further improve decision-making, we introduce Soft ACER, a novel loss function that utilizes confidence scores instead of binary labels, resulting in more stable training and better alignment with real-world evaluation metrics. TSANet achieves strong detection performance, with an average classification error rate 16.93 percent, outperforming state-of-the-art methods across combinations of various spoofing attacks and demonstrating resilience under severe class imbalance, and clean energy, improving livelihoods and advancing well-being across Africa.

Bio: Gabrial Zencha Ashungafac is a research associate at Carnegie Mellon University Africa and a member of the Vision and Language Intelligence Lab, working under Moise Busogi and Assane Gueye. His research focuses on computer vision, speech technologies, and bias quantification and mitigation in large language models, with particular emphasis on robustness, fairness, and generalization. He has contributed to projects on face anti-spoofing, robust object detection and classification, and 3D reconstruction, while also actively mentoring students in computer vision, deep learning, and applied machine learning. He holds a B.Sc. in computer engineering from TED University (Ankara, Turkey) and an M.Sc. in engineering artificial intelligence from Carnegie Mellon University Africa.

Biomedlink: A Web-Based Platform for Biomedical Equipment Management and AI-Assisted Technical Support

Speaker: Nzamubona Simeon, Ph.D. student, University of Rwanda, CEIoT

Abstract: In low-resource healthcare settings such as Rwanda, the effective management of biomedical equipment remains a significant challenge due to fragmented, paper-based maintenance systems and limited access to real-time technical support. This project proposes Biomedlink, an intelligent, web-based platform designed to enhance Health Technology Management (HTM) through automated scheduling, real-time project tracking, and AI-assisted technical support. Building on the success of the PharmaLab Web App—previously deployed to manage nationwide biomedical maintenance projects—Biomedlink aims to offer a scalable, multi-tenant solution aligned with international HTM and emerging intelligent HTM (iHTM) standards. The platform will incorporate preventive maintenance planning tools, Gantt chart visualizations, automated alerts, and an integrated AI chatbot trained on biomedical service manuals to support technicians in the field. Furthermore, Biomedlink will explore predictive maintenance capabilities using machine learning algorithms trained on historical service data. The project methodology includes system design and full-stack development using FastAPI, PostgreSQL, and modern web technologies, followed by pilot testing and evaluation with biomedical engineers in Rwanda and potentially Burundi. Expected outcomes include improved maintenance efficiency, reduced equipment downtime, enhanced technician support, and alignment with national eHealth transformation goals. The platform has already received interest from biomedical stakeholders in the region, highlighting its potential for commercialization and regional scale-up. Biomedlink represents a timely innovation in digital HTM, offering a blueprint for smarter, AI-driven biomedical equipment management in Africa and beyond.

Bio: Simeon Nzamubona is a physicist and biomedical engineer, pursuing a MSc in biomedical engineering at the University of Rwanda, College of Science and Technology. He is the chief of technical and biomedical services at PharmaLab Ltd, where he leads innovation in biomedical equipment management and healthcare technology. Simeon developed an AI-powered web application for Human Diagnostic Worldwide distributors, including Human Burundi and PharmaLab Ltd, which played a key role in managing a nationwide Rwanda Biomedical Center projects by improving efficiency, data-driven decision-making, and coordination across the country. He is also the developer of Biomedlink, a digital platform for biomedical project management and compliance, and also developer of TWIPLA, an initiative connecting STEM postgraduates with mentorship and funding. Passionate about advancing healthcare innovation in Africa, Simeon focuses on building impactful solutions at the intersection of physics, biomedical engineering, artificial intelligence, and entrepreneurship.

Upcoming Events