Md Adnan Arefeen is currently a 5th year CS PhD student in University of Missouri-Kansas City (UMKC) under the supervision of Dr. Md Yusuf Sarwar Uddin. Anticipating his graduation on January 13, 2025, Adnan is immersed in pioneering research aimed at revolutionizing the landscape of technology. At present, he is working a Research Intern at Integrated Systems department at NEC Labs America for Spring 2024 where he works on improving the retrieval augmented generation system (RAG) for videos. His journey with NEC commenced in Summer 2023 when as a research Intern, he developed a technology that reduces the cost of LLM API usage for domain-specific QA System. Adnan's scholarly pursuits extend across a diverse array of domains, including edge computing, efficient AI inference, video analytics, and cost-efficient LLM API utilization. His academic contributions shine brightly through his published research papers in prestigious conferences such as CVPR 2024, IoTDI 2023, DCOSS 2022, SMARTCOMP 2022-2023, ECML 2021, ICIP 2021, IEEE Big Data 2021. Adnan is from Bangladesh. Before joining the PhD program, he served as a lecture in the department of Computer Science and Engineering at United International University Bangladesh. He also served as a software engineer (iOS) at Reve Systems Ltd, Bangladesh.
The integration of Cyber-Physical Systems (CPS) with artificial intelligence (AI) has revolutionized various domains. notably enhancing surveillance systems like video analytics in autonomous vehicles. As surveillance cameras autonomously process visual data and detect anomalies, the demand for advanced AI-driven capabilities grows. However, the computational demands of deep neural network (DNN) models, particularly in computer vision (CV) and natural language processing (NLP), pose challenges for resource-constrained edge devices. Additionally, generative AI further challenges the CPS frameworks with very large language models (LLMs), making it almost impossible to run the entire application in resource-constrained edge devices. Moreover, with LLM API usage, it costs real currency to run video surveillance applications. Therefore, efficient processing of videos and other complex tasks with deep models and generative AI remains a challenge for CPS systems. Addressing this, this research focuses on finding solutions for these challenges.