Machine learning (ML)-assisted solutions for quality of transmission (QoT) estimation or classification have received significant attention in recent years. However, due to the unavailability of large and well-structured datasets, individual research groups need to create and use their own datasets for validating their proposed solutions. Therefore, the reported results (obtained using different datasets) are difficult to reproduce and hardly comparable. Regardless of this limitation, the unavailability of a technique to be followed by different research groups for the explainability of the dataset makes it even harder to validate the developed ML-assisted solutions across different papers. In this work, we present a publicly available dataset collection to open the problem of data-driven QoT estimation to the ML community. The dataset collection allows various solutions presented by different research groups to be compared. Furthermore, we present techniques to visualize and evaluate datasets for QoT estimation. The presented visualizations can also deliver deep insight into the error analysis of ML models. We apply these new methods to evaluate an artificial neural network on different datasets. The results show the relevance of the presented visualizations for comparing different approaches and different datasets. The proposed methods enable the comparison and validation of different ML-based solutions and published datasets.
@article{bergk2021mlassisted,author={Bergk, Geronimo and Shariati, Behnam and Safari, Pooyan and Fischer, Johannes K.},journal={Journal of Optical Communications and Networking},title={ML-assisted QoT estimation: a dataset collection and data visualization for dataset quality evaluation},year={2022},volume={14},month=mar,number={3},pages={43--55},keywords={Computational modeling;Data visualization;Data models;Optical fiber networks;Analytical models;Signal to noise ratio;Optical noise},doi={10.1364/JOCN.442733},}
Conference Proceedings
2024
Autonomous Capacity Adjustment with Dynamic Margin Allocation for Optical Enterprise Links
This work presents a novel machine learning-based dynamic capacity allocation scheme for efficient bandwidth provisioning of optical links. It offers an average hourly capacity saving of over 75% compared to traditional static capacity allocation mechanisms.
@inproceedings{balanici2024capacity,author={Balanici, Mihail and Shariati, Behnam and Safari, Pooyan and Bergk, Geronimo and Fischer, Johannes K.},booktitle={Optical Fiber Communication Conference (OFC) 2024},journal={Optical Fiber Communication Conference (OFC) 2024},keywords={Crosstalk; Energy; Machine learning; Optical networks; Phase},pages={M1H.2},publisher={Optica Publishing Group},title={Autonomous Capacity Adjustment with Dynamic Margin Allocation for Optical Enterprise Links},year={2024},url={https://opg.optica.org/abstract.cfm?URI=OFC-2024-M1H.2},doi={10.1364/OFC.2024.M1H.2},}
2022
Demonstration of a Real-Time ML Pipeline for Traffic Forecasting in AI-Assisted F5G Optical Access Networks
We showcase a proof-of-concept demonstration of a ML pipeline for real-time traffic forecasting deployed on a passive optical access network using an XGS-PON compatible telemetry framework. The demonstration reveals the benefits of fine-granular telemetry streaming for QoS monitoring and adaptive capacity adjustment of end-customers.
@inproceedings{Balanici:22,author={Balanici, Mihail and Bergk, Geronimo and Safari, Pooyan and Shariati, Behnam and Fischer, Johannes K. and Freund, Ronald},booktitle={2022 European Conference on Optical Communication (ECOC)},journal={2022 European Conference on Optical Communication (ECOC)},keywords={Machine learning; Networking hardware; Optical access networks; Optical network architecture; Optical networks; Passive optical networks},pages={Tu2.5},publisher={Optica Publishing Group},title={Demonstration of a Real-Time ML Pipeline for Traffic Forecasting in AI-Assisted F5G Optical Access Networks},year={2022},url={https://opg.optica.org/abstract.cfm?URI=ECEOC-2022-Tu2.5},}
Interactive Visual Analytics Dashboard for the Paradigm of ML-assisted Autonomous Optical Networking
We demonstrate a novel visualization dashboard, compatible with multiple data and telemetry sources, which offers dataset quality evaluation, dataset comparison, ML model error analysis interpretation, and network health monitoring.
@inproceedings{Shariati:22,author={Shariati, Behnam and Baltzer, Wanda and Bergk, Geronimo and Safari, Pooyan and Fischer, Johannes K.},booktitle={2022 Optical Fiber Communications Conference and Exhibition (OFC)},journal={2022 Optical Fiber Communications Conference and Exhibition (OFC)},keywords={Bit error rate; Machine learning; Modulation; Network topology; Optical networks; Optical signal to noise ratio},pages={M3Z.10},publisher={Optica Publishing Group},title={Interactive Visual Analytics Dashboard for the Paradigm of ML-assisted Autonomous Optical Networking},year={2022},url={https://opg.optica.org/abstract.cfm?URI=OFC-2022-M3Z.10},doi={10.1364/OFC.2022.M3Z.10},}
Traffic Monitoring and Analytics Framework for Optical Access Networks
This paper presents a Traffic Monitoring and Analytics (TMA) framework for optical access networks and defines the requirements and Key Performance Indicators (KPI) governing the telemetry collection, brokering, and streaming workflow as well as storage and computing resources of such a framework. Moreover, it introduces the concept of telecom data ownership and explores its impact on the realization of such a framework, primarily for the cases where equipment from multiple vendors co-exist in the operator networks. Eventually, considering the underlying requirements and the desired specifications, it presents several architectures for the realization of a TMA framework.
@inproceedings{shariati2022trafficmonitoring,author={Shariati, Behnam and Bergk, Geronimo and Safari, Pooyan and Balanici, Mihail and Fischer, Johannes and Freund, Ronald},booktitle={2022 13th International Symposium on Communication Systems, Networks and Digital Signal Processing (CSNDSP)},title={Traffic Monitoring and Analytics Framework for Optical Access Networks},year={2022},volume={},number={},pages={852--856},keywords={Optical communication equipment;Key performance indicator;Digital signal processing;Computer architecture;Optical fiber networks;Telecommunications;Telemetry;Telemetry;ML Pipeline;Traffic Prediction;Optical Access Networks},doi={10.1109/CSNDSP54353.2022.9907906},}
2021
Deep Convolutional Neural Network for Network-wide QoT Estimation
We propose a novel Deep Convolutional Neural Network formulation for network-wide QoT classification tasks and show its effectiveness for networks with significant topological differences. Our formulation achieves 99% accuracy on large and diverse test datasets.
@inproceedings{safari2021deepcnn,author={Safari, Pooyan and Shariati, Behnam and Bergk, Geronimo and Fischer, Johannes K.},booktitle={Optical Fiber Communication Conference (OFC) 2021},journal={Optical Fiber Communication Conference (OFC) 2021},keywords={Bit error rate; Erbium fibers; Network topology; Neural networks; Optical networks; Single mode fibers},pages={Th4J.3},publisher={Optica Publishing Group},title={Deep Convolutional Neural Network for Network-wide QoT Estimation},year={2021},url={https://opg.optica.org/abstract.cfm?URI=OFC-2021-Th4J.3},doi={10.1364/OFC.2021.Th4J.3},}
Inter-Operator Machine Learning Model Trading over Acumos AI Federated Marketplace
We demonstrate the development of a QoT classifier over an autonomous machine-learning pipeline, the trading of the classifier over a federated marketplace, and eventually its deployment in the customer’s network as a cloud-native micro-service.
@inproceedings{Shariati:21,author={Shariati, Behnam and Safari, Pooyan and Bergk, Geronimo and Oertel, Felix Immanuel and Fischer, Johannes K.},booktitle={Optical Fiber Communication Conference (OFC) 2021},journal={Optical Fiber Communication Conference (OFC) 2021},keywords={Crosstalk; Machine learning; Network topology; Neural networks; Optical networks; Statistics},pages={M2B.7},publisher={Optica Publishing Group},title={Inter-Operator Machine Learning Model Trading over Acumos AI Federated Marketplace},year={2021},url={https://opg.optica.org/abstract.cfm?URI=OFC-2021-M2B.7},doi={10.1364/OFC.2021.M2B.7},}
Secure Multi-Party Computation and Statistics Sharing for ML Model Training in Multi-domain Multi-vendor Networks
We propose a secure aggregation algorithm that allows proprietary-owned domains, hosting statistically different datasets, train and operate ML models in a Horizontally Federated Learning fashion. The obtained results show a compelling test accuracy of 98.60% for a QoT estimation use-case in multi-domain multi-vendor networks.
@inproceedings{safari2021secure,author={Safari, Pooyan and Shariati, Behnam and Bergk, Geronimo and Fischer, Johannes K.},booktitle={2021 European Conference on Optical Communication (ECOC)},title={Secure Multi-Party Computation and Statistics Sharing for ML Model Training in Multi-domain Multi-vendor Networks},year={2021},volume={},number={},pages={1--4},keywords={Training;Maximum likelihood estimation;Computational modeling;Biological system modeling;Europe;Collaborative work;Optical fiber communication},doi={10.1109/ECOC52684.2021.9606082},}
Book Chapters
Invited and professional contributions
2025
Einkauf–Optimierung entlang des Einkaufsprozesses mit KI und Daten
Benito Cervellera, Geronimo Bergk, and Hendrik Schlünsen
Die fortschreitende Digitalisierung sowie steigende Anforderungen durch komplexere Lieferketten, volatile Märkte und höhere Regulierungen stellen neue Herausforderungen für den Einkauf dar. Künstliche Intelligenz (KI) bietet erhebliche Potenziale für Optimierungen entlang des Einkaufsprozesses. KI ermöglicht z. B. Prozesse zu automatisieren, strategische Entscheidungen datenbasiert zu treffen und Effizienzpotenziale zu realisieren, etwa durch Szenarioanalysen, Unterstützung des Ausschreibungsprozesses oder die Automatisierung von Routineaufgaben. KI unterstützt bereits heute bei der Erstellung von Ausschreibungsunterlagen, der Auswertung komplexer Angebote sowie der Analyse von Vertragsklauseln für die Vorbereitung von Vertragsverhandlungen. Ein zentraler Erfolgsfaktor für die Integration von KI im Einkauf ist die Sicherstellung einer hohen Datenqualität sowie die Anpassung organisatorischer Strukturen, um datenbasierte Entscheidungsfindung zu fördern. Zugleich ist die Schulung von Mitarbeitern sowie die Förderung einer offenen Unternehmenskultur notwendig, um Akzeptanz und Vertrauen in KI-gestützte Lösungen zu schaffen. Die Transformation erfordert eine klare Strategie, die technologische und kulturelle Hürden adressiert, um langfristig Effizienz, Agilität und Innovationskraft im Einkauf zu steigern.
@inbook{cervellera2025performanceintelligence,title={Einkauf–Optimierung entlang des Einkaufsprozesses mit KI und Daten},author={Cervellera, Benito and Bergk, Geronimo and Schl{\"u}nsen, Hendrik},booktitle={Performance Intelligence},publisher={Springer Fachmedien Wiesbaden},year={2025},language={german},}
