nmrg

Summary

This meeting of the NMRG focused on several key topics: IBM use case consolidation, energy-aware security research directions, autonomic control loops for network slice reconfiguration, and data quality challenges in machine learning-based intrusion detection systems (IDS). The group discussed the adoption of a consolidated IBM use case draft, explored future research on energy-aware security, and examined methods for autonomic network slice reconfiguration using AI and NLP. A significant portion of the meeting was dedicated to the challenges of data quality and evaluation in machine learning-based IDS, with presentations on improving test set quality and assessing the impact of data quality on model performance.

Key Discussion Points

IBM Use Case Draft: Discussion around merging multiple IBM-related drafts into a single document. Concerns were raised about individual draft development and the need for clarity on the content included in the collective document.
Energy-Aware Security: Presentation on future research directions for energy-aware security mechanisms, focusing on measuring the energy consumption of security protocols and the trade-offs between security, performance, and energy usage. The right venue for this research within the IETF/IRTF was debated.
Autonomic Control Loop for Network Slicing: Presentation on using an autonomic control loop with automated planning and NLP to achieve self-reconfiguration in sliced networks. The discussion focused on the translation of natural language intents into machine-understandable instructions.
Data Quality in IDS: Three presentations highlighted issues with data quality and evaluation in machine learning-based IDS. The discussions covered topics such as overfitting, the lack of variability and complexity in existing datasets, and methods for improving test set quality.
Test Set Quality and Realism: The challenges in creating realistic test sets for IDS were emphasized. The limitations of lab-based evaluations and the need for better datasets representative of real-world traffic were discussed.
Model Interpretability: The importance of model interpretability for understanding the behavior of machine learning-based IDS and identifying potential data quality issues was highlighted.

Decisions and Action Items

IBM Use Case Draft Adoption: Proceed with a call for adoption of the consolidated IBM use case draft on the mailing list. Clarify the status of Luis's individual drafts before the IRG review.
Jefferson's Energy-Aware Security Draft: Jefferson to consider incorporating research questions into his draft and assess the best venue for the work (NMERG or other groups).
Data Quality in IDS: The group acknowledged the need for better datasets and evaluation methodologies for machine learning-based IDS.

Next Steps

The co-chairs will initiate the call for adoption of the IBM use case draft on the mailing list.
Jefferson will refine his energy-aware security draft and consider the appropriate venue for further work.
The group will continue the discussion on AI network management during the Friday session.
Discussions on data quality and evaluation methods in IDS will continue on the mailing list and in future meetings.

Session Date/Time: 08 Nov 2024 15:30

nmrg

Summary

This meeting of the Network Management Research Group (NMRG) covered document status updates, presentations on AI in network management, updates on the CLASS architecture, and a review of the research agenda. Key discussions centered on the application of AI in network management, the architecture of AI-based agents, and the evolution of network digital twin concepts. The meeting concluded with updates to the research group's research agenda, highlighting key areas of focus for future work.

Key Discussion Points

AI-Based Internet Management Agent (Shin): Presentation on a draft proposing a common architecture for AI in network management, specifically in the form of a network management agent (NMA). Discussion covered the independence vs. integration of the NMA with existing network management systems, and the need to clearly define the role of AI and its differentiation from other management approaches. Concerns were raised regarding potential conflicts and the need for shared data and models.
Distributed AI Systems for Network Management (Pedro): Presentation on constructing distributed AI systems, focusing on seamless integration among elements using a publish-subscribe system. The discussion centered on the specific use cases, the distributed nature of services across multiple administrative domains, and the relationship to AI orchestration pipelines. It was noted that the work could potentially be complementary to the NMA work previously presented.
Considerations of Network Systems for AI Services (Yongkun): Presentation on a draft addressing challenges in coupling AI and network management. The presentation outlined how the draft addresses low-level and high-level challenges identified in previous research. Discussion focused on the potential overlap with Pedro's work and the need to carefully word claims of addressing challenges. The draft was confirmed to be adopted as a research group document and will be made more general following the suggestions.
CLASS Architecture Update (Luis): Presentation on an evolution of the CLASS architecture to include a compute stratum and a knowledge plane. Discussion focused on the motivation behind the architecture, the interplay between cloud and network infrastructures, and the sharing of information between different domains. Questions arose regarding the applicability of the work to energy.
Network Digital Twin (NDT) Update (Shin): Update on the Network Digital Twin draft, highlighting the resolution of open issues and the alignment with ETSI's work on NDT. Key changes include the addition of a "logic" element and the redefinition of the architecture as a conceptual architecture rather than a functional one.
Research Agenda: Review of the research agenda including themes of AI & Network Management, Data-Centricity, and Network Digital Twins.

Decisions and Action Items

Consideration of Network Systems for AI Services: The document will be adopted as a research group document.
Net Digital Twin Concept and Reference Architecture: Last call to be initiated on the document.
Challenges and Opportunities in Management for Green Networking: Is in IR pole.
Research challenge in coupling artificial intelligence and network management: To be submitted to RISU review.
Use cases and practices for inter-based networking: To be opened for call for adoption on the mailing list.

Next Steps

Authors of AI-based Internet Management Agent (Shin) to address feedback and further refine the architecture.
Authors of Considerations of Network Systems for AI Services (Yongkun) and Distributed AI Systems for Network Management (Pedro) to coordinate their work and identify potential areas of collaboration.
Authors of CLASS Architecture Update (Luis) to consider feedback and explore potential use cases within the context of NMRG.
Network Digital Twin to undergo group last call.
Update and Publish the Research Agenda.