Expected Results

By the end of this project's implementation, the following key results are to be expected:

• Quasi-automatic MLOps pipeline responsible for deployment of better-performing models: The system will be able to automatically deploy and roll back Machine Learning (ML) models from the system's usage, although manual intervention could be marginally required. While a model is active, other models may be continuously training in parallel until one's performance thrives above thresholds beyond the active model's; thus the need for such an automated workflow.

• Integration of the MLOps pipeline with a network: The system should be able to partake in communications on a network regardless of its native context (5G, 4G...) for the receiving and retrieving of data.

• Automatic data ingestion, governance, and processing, given network metrics: These metrics (which can be sourced, for example, from anything IoT-related to industrial machinery) will enter the system for further training, processing, evaluation, and security-based aspects.

• Administrative dashboard for system status monitoring and maintenance: All-in-one interface to ensure efficient system management, while also able to be role-specific to restrict access to different user types.

• Risk/Security assessments for pipeline-specific decisions: Human-in-the-loop and/or automated risk management, in which the latter should be the main actor with most of the uptime in this task. Decisions created by the system's active machine learning model will be categorized in terms of risk and be constantly mediated to look out for improper courses of action.

These expected results will grant a flexible system with secure communication, holistic access to administrators for maintainability, and an MLOps pipeline responsible for the training and integration of increasingly better models for the desired context. Therefore, upbringing smarter networks, sensitive to the environment data that flows into the processing, storage, and filtering.