At the Security, Privacy, and Intelligence for Edge Devices (SPEED) Lab , our vision is to drive transformative research in the domain of AI, federated learning, large language models, vision-language models, cybersecurity, and multi-modality. We strive to place strong emphasis on edge intelligence, tailoring the potential of distributed computing to enable efficient decision-making. We aspire to pioneer advancements that redefine the boundaries of AI and machine learning, while ensuring the security and privacy of distributed systems. Through our interdisciplinary research, we strive to develop robust, efficient, and intelligent solutions that enhance decision-making in edge environments, driving innovation that positively impacts society in an increasingly interconnected world.
NSF CRII Grant of $167,500 Awarded to Dr. Imteaj! Thanks to NSF!
Dr. Imteaj’s project with FIU won nearly $100,000 DHS grant!
Zarif presented paper at IEEE SmartComp!
Saika won the Graduate Research Prize at Tapia'23 Conference.
Dr. Imteaj received the Prestigious FIU Real Triumph Award!
Best Paper Award at IEEE CSCI conference!
Dr. Ahmed Imteaj Named in Stanford’s Top 2% Scientists for Research Impact
Anouncement: A fully funded Ph.D. position is available! We are looking for strongly motivated candidates having experience in Large Language Models, Federated Learning or Computer Vision. We are always looking forward to work with SIUC undergraduated and MS students.
Our project focuses on developing visual language model that can accurately interpret and generate natural language descriptions for visual content. Through innovative techniques, we aim to create a model that not only improves the understanding of visual data but also enhances the generation of meaningful and contextually relevant textual descriptions. This research has applications in image captioning, visual question answering, and various other tasks that require a deep understanding of both visual and textual data.
This research delves into the realm of crime detection through a pioneering integration of drone technology and developing a novel keypoint detection algorithm. The ongoing project focuses on the acquisition of criminal activity data using drones and the application of advanced keypoint detection algorithms to decipher intricate patterns and anomalies. The overarching goal of this initiative is to provide law enforcement agencies with a powerful tool to elevate their situational awareness and proactively address criminal behavior, leveraging a unique aerial perspective. By combining the capabilities of drones and cutting-edge algorithms, this project seeks to contribute to the advancement of crime detection methodologies, offering law enforcement a strategic advantage in identifying and responding to criminal activities with greater efficiency and precision.
Federated Meta-Learning (FML) can produce personalized model with limited training data which is suitable for resource-constrained edge devices. Nonetheless, the FML process may susceptible to adversarial attacks. This project delves into the dynamic landscape of FML under adversarial attacks and propose a novel FML algorithm, FLAMINGO that enhance meta-learners robustness against adversarial attacks and prevent overfitting through adversarial meta-training and consistency regularization, all while minimizing communication cost.
The aim of this project is to promptly identify and respond to potential life-threatening criminal activities and safety concerns. WatchOverGPT combines the capabilities of wearable cameras, smartphones’ location data, and large language model (LLM)-based advanced conversational AI communication through ChatGPT. Through real-time data processing and AI-driven autonomous communication, the framework aims to bridge the gap between safety threats and emergency response teams in the fight against crime and enhance the security of individuals in various settings
Our research focus revolves around developing cutting-edge Robust, Lightweight, and Secure Distributed Machine Learning Algorithms. Emphasizing robust intelligence, our algorithms dynamically adapt to diverse data distributions, ensuring consistent performance in dynamic settings. Prioritizing lightweight solutions, our goal is to enhance efficiency without compromising accuracy, ideal for resource-constrained environments. Exploring the realm of securing distributed learning, our research aims to establish a robust foundation for trustworthy and resilient machine learning applications in diverse and evolving scenarios.
This project leverages federated learning to analyze aerial images obtained via drone surveillance. Through collaborative learning across individual drones, the project aspires to develop a sophisticated system capable of real-time analysis, precise identification, and classification of land cover and land use patterns. The envisioned system holds vast potential for applications across diverse domains including environmental monitoring, urban planning, agriculture, disaster response, and beyond.
This project pioneers a cutting-edge approach to traffic management by leveraging Federated Learning and Social Media Analysis for real-time insights. The model refines its understanding of dynamic traffic patterns without centralized data aggregation. The integration of social media data provides contextual information, enriching the accuracy of real-time traffic assessments. Ultimately, the project envisions a smarter road network with real-time analytics for efficient traffic management.
The responsible development and deployment of Generative AI (Artificial Intelligence) is a critical imperative as this technology continues to advance. Generative AI, particularly exemplified by models like GPT (Generative Pre-trained Transformer), has demonstrated remarkable capabilities in generating human-like text, images, and more. However, with great power comes great responsibility. Our focus in responsible development lies in ethical considerations, transparency, and mitigating potential risks.
This project is dedicated to develop a tailored Distributed Machine Learning Framework designed specifically for interconnected Cyber-Physical-Societal Networks. Our focus is on enabling interdependent decision-making processes by extracting profound insights from the interconnected networks. This framework aims to comprehend the complex relationships between these networks, facilitating a deeper understanding of their interdependence and empowering informed decision-making for enhanced resilience and efficiency across interconnected systems.
This project involves orchestrating a fleet of DJI Tello drones to operate collectively. Through distributed streaming, these drones exchange real-time data and insights. This data sharing facilitates a collaborative learning environment where each drone contributes its observations and experiences to a shared knowledge base. This collective intelligence enables the swarm to exhibit synchronized behavior and adaptive decision-making, enhancing their ability to navigate complex environments, perform tasks efficiently, and respond dynamically to changing scenarios.