Modern machine learning (ML) systems require robust pipelines to ensure reproducibility, scalability, and continuous deployment. MLOps integrates data engineering, model training, validation, deployment, and monitoring into a unified automated workflow. This thesis focuses on developing an end-to-end MLOps demonstrator that automates the full ML lifecycle – from edge data acquisition to cloud-based training and deployment back to edge devices.
The demonstrator will highlight state-of-the-art concepts such as CI/CD for ML, transfer learning, automated retraining, and edge model updates. The result will be a practical proof of concept showcasing continuous ML operations in distributed environments.
Student Target Groups:
- Students of ICE/Telematics;
- Students of Computer Science;
- Students of Software Engineering.
Thesis Type:
- Bachelor Thesis / Master Project / Master thesis
Goal and Tasks:
The primary goal of this thesis is to design and implement an end-to-end MLOps pipeline that automates data flow, model training, evaluation, deployment, and updating at edge devices.
- Conduct a state-of-the-art research review on MLOps frameworks, CI/CD for ML, and edge-cloud ML architectures;
- Set up a data pipeline that transmits edge-generated data securely to a cloud-based processing environment;
- Implement ML model training and retraining processes, including transfer learning techniques;
- Research mechanisms for automated ML model deployment and updates on edge devices;
- Validate trained ML models using defined performance metrics (e.g., accuracy, precision, recall, latency, robustness);
- Document the architecture, implementation process, evaluation results, and lessons learned;
- Prepare and deliver an oral presentation summarizing the results and conclusions.
Recommended Prior Knowledge:
- Basic knowledge of machine learning;
- Programming experience (e.g., Python);
- Basic understanding of cloud and networking concepts;
- Interest in the topic.
Start:
Contact: