Scope of the Workshop
Communication networks have evolved drastically in the last decade. Whereas networks largely used to provide dumb connectivity pipes interconnecting its end users, current network technology is tightly interconnected with the cloud, leading to a plethora of advanced services, a drastic increase in network usage, and strongly evolved data and control planes. Software-based functionality is now deeply changing the nature of both the control and data plane of our networking infrastructure through SDN and NFV technology respectively. This has introduced tremendous programmability and flexibility but also a range of uncertainties in the performance, security and management of our networks. Less functionality is now specified in standardized protocols or hardcoded in our data plane hardware.
Until recently, machine learning techniques have been used in networking mostly for mechanisms outside of the control loop or fast path of our networks. Whereas machine learning has been effectively used for anomaly detection, network prediction or analytical purposes, the increasing network softwarization is now creating potential for apply recent evolutions in machine learning to optimize the actual control and data plane operation of networking infrastructure. Such applications might, for example, learn and optimize network performance relationships between softwarized network functions, presented loads and potential software/hardware configurations. The increasing availability of monitoring data and associated monitoring platforms might even further fuel advanced ML techniques including deep learning in exploiting ML-driven network approaches.
We are hence interested in papers showing how ML can improve and benefit from softwarized networks, e.g., by:
- Solving typical algorithmic problems that appear in such a network, typically in a MANO component (e.g., scaling, placement, admission control, load-dependent flow routing, …),
- Taking actual decisions about the operation of a network, e.g., network resource allocation in a wireless network, computing MAC schedules or taking handover decisions
- Summarize observational data of a network, e.g., compressing monitoring data to meaningful results or deciding where in a network monitoring is required to detect inconsistent behavior
- Supporting the developer of network functions or chains by dealing with performance data, building profile information out of individual measurements
We are also interested in the other direction: how can an ML scheme be suitably supported by the abilities of a softwarized network where arbitrary functions (not just network functions) can be distributed into the network and executed locally. This can entail:
- Distributed machine learning schemes
- Transfer learning from one part of a network towards another (e.g., to leverage time-of-day usage patterns around the globe, or from one network installation to another)
- Distributed storage of input/observational data
- Continuous learning and retraining in a distributed setting
This workshop is a satellite workshop of the NetSoft 2020 conference June 29-July 3, 2020, in Ghent, Belgium. All guidelines and deadlines of the NetSoft workshops apply here as well; please compare there.
- Wouter Tavernier, Ghent University & imec
- Holger Karl, Paderborn University
Technical Program Committee