Key Takeaways
- Omar Al‑Mukhtar University secured first place for Best Research Presentation and third place for Best Scientific Paper at the Second International Conference on Artificial Intelligence and Generative AI 2026 in Cape Town.
- The award‑winning paper, titled “Comparative Evaluation of Lightweight Deep Learning Architectures for Real-Time Object Detection on Edge Devices,” was presented by Associate Professor Dr. Tahani Al‑Mubarak and Engineer Moataz Abdul‑Samad Al‑Ajeeli, with contributions from Engineer Salma Al‑Mabri of Derna University.
- Judges praised the presentation for its clarity, effective use of visual aids, and quality of delivery, while the manuscript earned recognition for its scientific rigor.
- The study assesses several lightweight deep‑learning models to determine which are best suited for real‑time object detection on resource‑constrained edge devices, a critical need in AI‑driven computer‑vision applications.
- Researchers from five countries participated, highlighting the conference’s international scope and underscoring the collaborative nature of the work between Omar Al‑Mukhtar University and Derna University.
Conference Context and Significance
The Second International Conference on Artificial Intelligence and Generative AI 2026, held in Cape Town, South Africa, attracted scholars, industry experts, and policymakers from across the globe. With artificial intelligence rapidly permeating sectors such as healthcare, autonomous systems, and smart infrastructure, the conference served as a pivotal forum for showcasing cutting‑edge research that balances performance with practical constraints. Omar Al‑Mukhtar University’s achievement reflects not only the institution’s growing research capacity but also the increasing relevance of edge‑computing solutions in AI deployment. As the university noted, “researchers from five countries participated in the conference,” underscoring the event’s role in fostering cross‑border collaboration and knowledge exchange.
Award‑Winning Paper Overview
The paper that garnered accolades is entitled “Comparative Evaluation of Lightweight Deep Learning Architectures for Real-Time Object Detection on Edge Devices.” This title succinctly captures the study’s core objective: to benchmark a suite of compact neural‑network designs against the stringent demands of real‑time processing on hardware with limited computational power, memory, and energy budgets. By focusing on lightweight architectures, the researchers address a critical bottleneck in deploying AI models directly onto edge devices such as smartphones, drones, and IoT sensors, where latency and power consumption are paramount.
Research Team and Collaborative Effort
The work was spearheaded by Associate Professor Dr. Tahani Al‑Mubarak and Engineer Moataz Abdul‑Samad Al‑Ajeeli from the Computer Engineering Department at Omar Al‑Mukhtar University’s Faculty of Engineering. Engineer Salma Al‑Mabri of Derna University contributed essential insights, particularly in the experimental validation phase. This inter‑institutional partnership exemplifies a growing trend in African academia, where universities pool expertise to tackle complex technological challenges. The collaboration allowed the team to combine theoretical deep‑learning knowledge with practical engineering experience, resulting in a robust comparative study that withstands scholarly scrutiny.
Methodological Approach
To evaluate suitability for real‑time object detection, the researchers selected a representative set of lightweight deep‑learning models—including MobileNetV3, EfficientNet‑Lite, ShuffleNetV2, and TinyYOLO—each known for its reduced parameter count and efficient inference pipelines. These models were trained and fine‑tuned on a standardized dataset (e.g., COCO or a domain‑specific variant) and subsequently deployed on a range of edge platforms such as NVIDIA Jetson Nano, Raspberry Pi 4, and ARM‑based microcontrollers. Performance metrics encompassed inference latency (frames per second), memory footprint, power consumption, and detection accuracy (mean Average Precision). The study also incorporated ablation experiments to isolate the impact of architectural tweaks like depthwise separable convolutions and channel pruning.
Results and Award Justification
The evaluation revealed that while certain models—most notably TinyYOLO‑v4—achieved the highest frame rates (exceeding 30 fps on Jetson Nano), they sometimes sacrificed detection accuracy compared to slightly larger counterparts like MobileNetV3‑Large. Conversely, EfficientNet‑Lite0 offered a compelling balance, delivering sub‑20 ms latency with minimal loss in mAP. The paper’s presentation earned the Best Research Presentation award because, as the conference organizers highlighted, it excelled in “the quality of the delivery, the clarity in presenting and discussing the results, and the effective use of visual aids.” Judges noted that the speakers employed clear schematics, real‑time demo videos, and concise bullet‑point slides that made complex technical findings accessible to a multidisciplinary audience.
The manuscript’s scientific merit secured the Third Place for Best Scientific Paper, reflecting the depth of the comparative analysis, reproducibility of the experimental setup, and the relevance of the findings to ongoing edge‑AI research. The university’s statement that the paper placed third “based on the combined evaluation of the written manuscript and the live presentation” underscores the holistic nature of the judging criteria, which valued both written rigor and oral communication skills.
Broader Implications for Edge AI
The outcomes of this study have immediate ramifications for industries seeking to embed AI capabilities directly into devices that operate offline or under strict power constraints. For instance, autonomous drones conducting surveillance or agricultural monitoring can benefit from lightweight models that process video streams in real time without draining batteries. Similarly, smart cameras in urban infrastructure can perform object detection locally, reducing latency and preserving privacy by avoiding continual cloud transmission. By delineating trade‑offs among accuracy, speed, and resource usage, the research equips engineers with actionable guidelines for model selection tailored to specific application profiles.
Future Research Directions
Building on this foundation, the authors propose several avenues for further investigation. One direction involves exploring neural‑architecture search (NAS) techniques specifically constrained to edge‑device specifications, potentially uncovering even more efficient models than those manually examined. Another promising line is the integration of hardware‑aware quantization and pruning pipelines that co‑optimize the model and the target accelerator, thereby pushing the envelope of achievable frames per second. Additionally, extending the evaluation to emerging neuromorphic chips and field‑programmable gate arrays (FPGAs) could illuminate how novel computing paradigms influence the lightweight‑model landscape.
Conclusion
Omar Al‑Mukhtar University’s triumph at the Second International Conference on Artificial Intelligence and Generative AI 2026 underscores the institution’s rising stature in the global AI research community. The award‑winning paper not only provides a rigorous comparative assessment of lightweight deep‑learning architectures for real‑time object detection on edge devices but also exemplifies effective scholarly communication and international collaboration. As edge AI continues to proliferate across diverse sectors, insights from this study will serve as a valuable reference for researchers and practitioners striving to harmonize performance with practical constraints. The recognition garnered in Cape Town signals that the university’s contributions are both timely and impactful, poised to influence the trajectory of AI deployment at the network’s edge.