Nokia Bell Labs proposes learning life-cycle for faster adoption of autonomic optical networks

Espoo, Finland – Nokia Oyj's research division, Nokia Bell Labs, has unveiled a new research paper focused on enhancing the implementation of autonomic optical networks. The publication introduces a learning life-cycle framework intended to speed up the adoption of autonomous optical transmission and networking technologies.

The paper highlights a key challenge: machine learning (ML) models require substantial datasets for accurate training. However, the continuous deployment of new optical equipment and techniques in live networks often limits the availability of sufficient real-world data. Data generated from simulations and lab experiments may not fully capture the complexity and variety of operational environments, leading to potential inaccuracies in ML models.

Nokia Bell Labs proposes an ML-based algorithm life-cycle to facilitate the deployment and ongoing refinement of these models within real networks. The approach suggests initial model training using data from simulations and lab experiments. Once deployed, models can be re-trained and fine-tuned based on detected inaccuracies, thereby improving their precision over time. The research illustrates the benefits of this learning cycle with numerical results and presents two specific use cases.

One demonstrated use case involves a two-phase strategy: initial out-of-field training with generic data, followed by in-field adaptation to accommodate diverse equipment. This method proved effective for failure detection and identification. Another use case focuses on in-field re-training, where ML models are updated dynamically after performance degradation is observed, showcasing significant advantages through collective learning for autonomic transmission.