Comparative Analysis of Edge AI Architectures for Real-Time IoT Applications

Thesis Abstract

The rapid proliferation of Internet of Things (IoT) devices has underscored the critical need for efficient edge artificial intelligence (AI) architectures capable of delivering real-time data processing and decision-making while minimizing latency and conserving bandwidth. Despite advancements in edge computing, the diversity in architectural designs—such as cloud-edge hybrid, embedded, and specialized AI accelerator frameworks—presents challenges in identifying the most suitable architecture for various IoT applications, particularly those demanding high reliability and low latency. This study aims to conduct a comprehensive comparative analysis of prominent edge AI architectures to determine their performance, scalability, energy efficiency, and suitability for real-time IoT scenarios. Specifically, the research objectives include evaluating computational latency, power consumption, and accuracy across different architectural frameworks, as well as identifying optimal deployment strategies aligned with specific IoT application requirements. Employing a quantitative research design, the study utilizes a mixed-methods approach to gather empirical data. The population under investigation comprises 15 edge AI platforms and frameworks, including NVIDIA Jetson Nano, Google Coral, Intel Movidius Neural Compute Stick, and custom embedded systems, selected based on their prominence in real-time IoT deployments. A stratified sampling technique ensures the inclusion of diverse architectures across hardware, software, and scalability dimensions. Data collection instruments consist of standardized benchmarking tools—such as the Edge AI Benchmark Suite—and custom scripts to measure latency, throughput, power consumption, and accuracy in controlled laboratory environments simulating various IoT application contexts like smart surveillance, predictive maintenance, and autonomous vehicles. Data analysis employs descriptive statistics to summarize performance metrics and inferential statistical techniques, notably analysis of variance (ANOVA), to detect significant differences among architectures across evaluated parameters. Complementary regression analyses explore correlations between hardware specifications and performance outcomes. Qualitative data from expert interviews are subjected to thematic analysis, providing insights into deployment challenges and architectural preferences. The study applies theoretical frameworks such as the Technology Acceptance Model (TAM) and Distributed Artificial Intelligence theory to interpret findings related to user adoption and system scalability. Expected results anticipate delineating clear performance hierarchies among the architectures, with embedded AI frameworks generally demonstrating lower latency and power consumption but potentially limited scalability, while cloud-edge hybrid systems offer enhanced scalability at the expense of increased latency. Findings are projected to reveal critical trade-offs and design considerations, guiding practitioners and developers in selecting suitable architectures tailored to diverse real-time IoT applications. Additionally, the study aims to identify gaps in current architectural designs, particularly regarding resource optimization and interoperability, fostering avenues for future innovation. The contribution to knowledge resides in providing a systematic, data-driven comparison of edge AI architectures, filling existing literature gaps concerning practical performance metrics in real-world IoT contexts. It also proposes a decision matrix framework that aligns architectural choices with specific application demands and resource constraints. The main conclusion underscores that no singular architecture universally outperforms others; instead, context-dependent factors such as application criticality, resource availability, and scalability needs should inform deployment strategies. Based on these insights, recommendations include establishing standardized benchmarking protocols for edge AI, integrating adaptive architecture selection mechanisms, and fostering interdisciplinary collaboration for enhanced system design. Future research suggestions emphasize longitudinal field studies and exploring emerging architectures incorporating neuromorphic computing and federated learning paradigms to further advance the domain of real-time IoT edge AI systems.