Abstract
TinyML is an evolving field in wireless communication and edge intelligence. This thesis integrates machine learning models into edge devices to monitor and predict real-time network usage, aiming to optimize wireless network utilization via dynamic monitoring and forecasting. A hybrid CNN-LSTM model was selected to capture spatial and temporal dependencies, achieving high accuracy (95%) in traffic prediction. This thesis also explores GPT and LLM technologies in network monitoring, discussing practical applications and energy-efficient strategies.
Objectives
- Develop a CNN-LSTM model to forecast network traffic accurately.
- Analyze the challenges of deploying AI models on edge devices.
- Create a real-time network monitoring application with forecasting capability.
Research Questions
- How accurately can a CNN-LSTM model forecast network traffic?
- What are the challenges of deploying generative AI models on edge devices?
Contributions
- Demonstrates feasibility of ML models on edge devices for low-latency applications.
- Design and deployment of an edge-enhanced monitoring system with traffic forecasting.
- Insights into applying GPT and LLM technologies in network monitoring.
- Development of a dockerized application for real-time monitoring.
- Focus on energy-efficient ML practices on edge devices.
Conclusion
This thesis advances wireless communication and edge device optimization, contributing to machine learning's potential in real-time applications. Future research may explore hybrid ML approaches, improving network management and real-time forecasting on edge devices.
