arXiv subjects

Nuria Molner

Publications and source records attributed to Nuria Molner.

7 recordsLinked to original sources

arXiv2025

AIORA: An AI-Native Multi-Stakeholder Orchestration Architecture for 6G Continuum

This paper elaborates on a novel AI-native architecture for emerging 6G systems harnessing open APIs, along with supporting mechanisms to empower intelligent and coordinated orchestration of edge-cloud continuum resources. The AIORA architecture facilitates a seamless creation, life-cycle management, and exposure of services in multi-segment heterogeneous environments. It integrates new breeds of tools and advanced technologies to enable zero-touch management of an edge-cloud continuum, building on top of the 3GPP Edge Enablement Layer and the respective connectivity models, allowing to cater to the high flexibility, availability, efficiency, reliability, and resilience needs of the future 6G services and applications. Several ongoing industry initiatives -- such as ETSI MEC for edge computing platforms, the GSMA Operator Platform for multi-operator service federation, and CAMARA for cross-operator API standardization -- demonstrate the growing momentum towards integrated frameworks where edge, cloud, and network resources can be seamlessly orchestrated. Our proposed AIORA architecture not only aligns with these initiatives but also extends them by leveraging a multi-segment virtual continuum concept and nested AI-driven closed loops for real-time optimization.

cs.NI↗

arXiv2024

Combining Relevance and Magnitude for Resource-Aware DNN Pruning

Pruning neural networks, i.e., removing some of their parameters whilst retaining their accuracy, is one of the main ways to reduce the latency of a machine learning pipeline, especially in resource- and/or bandwidth-constrained scenarios. In this context, the pruning technique, i.e., how to choose the parameters to remove, is critical to the system performance. In this paper, we propose a novel pruning approach, called FlexRel and predicated upon combining training-time and inference-time information, namely, parameter magnitude and relevance, in order to improve the resulting accuracy whilst saving both computational resources and bandwidth. Our performance evaluation shows that FlexRel is able to achieve higher pruning factors, saving over 35% bandwidth for typical accuracy targets.

cs.LG↗

arXiv2023

Choose, not Hoard: Information-to-Model Matching for Artificial Intelligence in O-RAN

Open Radio Access Network (O-RAN) is an emerging paradigm, whereby virtualized network infrastructure elements from different vendors communicate via open, standardized interfaces. A key element therein is the RAN Intelligent Controller (RIC), an Artificial Intelligence (AI)-based controller. Traditionally, all data available in the network has been used to train a single AI model to be used at the RIC. This paper introduces, discusses, and evaluates the creation of multiple AI model instances at different RICs, leveraging information from some (or all) locations for their training. This brings about a flexible relationship between gNBs, the AI models used to control them, and the data such models are trained with. Experiments with real-world traces show how using multiple AI model instances that choose training data from specific locations improve the performance of traditional approaches following the hoarding strategy.

cs.NI↗

arXiv2022

Network Support for High-performance Distributed Machine Learning

The traditional approach to distributed machine learning is to adapt learning algorithms to the network, e.g., reducing updates to curb overhead. Networks based on intelligent edge, instead, make it possible to follow the opposite approach, i.e., to define the logical network topology em around the learning task to perform, so as to meet the desired learning performance. In this paper, we propose a system model that captures such aspects in the context of supervised machine learning, accounting for both learning nodes (that perform computations) and information nodes (that provide data). We then formulate the problem of selecting (i) which learning and information nodes should cooperate to complete the learning task, and (ii) the number of iterations to perform, in order to minimize the learning cost while meeting the target prediction error and execution time. After proving important properties of the above problem, we devise an algorithm, named DoubleClimb, that can find a 1+1/|I|-competitive solution (with I being the set of information nodes), with cubic worst-case complexity. Our performance evaluation, leveraging a real-world network topology and considering both classification and regression tasks, also shows that DoubleClimb closely matches the optimum, outperforming state-of-the-art alternatives.

cs.NI↗

arXiv2020

Delay and reliability-constrained VNF placement on mobile and volatile 5G infrastructure

The ongoing research and industrial exploitation of SDN and NFV technologies promise higher flexibility on network automation and infrastructure optimization. Choosing the location of Virtual Network Functions is a central problem in the automation and optimization of the software-defined, virtualization-based next generation of networks such as 5G and beyond. Network services provided for autonomous vehicles, factory automation, e-health and cloud robotics often require strict delay bounds and reliability constraints, which are influenced by the location of its composing Virtual Network Functions. Robots, vehicles and other end-devices provide significant capabilities such as actuators, sensors and local computation which are essential for some services. Moreover, these devices are continuously on the move and might lose network connection or run out of battery power, which further challenge service delivery in this dynamic environment. This work tackles the mobility, and battery restrictions; as well as the temporal aspects and conflicting traits of reliable, low latency service deployment over a volatile network, where mobile compute nodes act as an extension of the cloud and edge computing infrastructure. The problem is formulated as a cost-minimizing Virtual Network Function placement optimization and an efficient heuristic is proposed. The algorithms are extensively evaluated from various aspects by simulation on detailed real-world scenarios.

cs.NI↗

arXiv2018

Arbitration Among Vertical Services

A 5G network provides several service types, tailored to specific needs such as high bandwidth or low latency. On top of these communication services, verticals are enabled to deploy their own vertical services. These vertical service instances compete for the resources of the underlying common infrastructure. We present a resource arbitration approach that allows to handle such resource conflicts on a high level and to provide guidance to lowerlevel orchestration components.

cs.NI↗

arXiv2018

Resource Orchestration of 5G Transport Networks for Vertical Industries

The future 5G transport networks are envisioned to support a variety of vertical services through network slicing and efficient orchestration over multiple administrative domains. In this paper, we propose an orchestrator architecture to support vertical services to meet their diverse resource and service requirements. We then present a system model for resource orchestration of transport networks as well as low-complexity algorithms that aim at minimizing service deployment cost and/or service latency. Importantly, the proposed model can work with any level of abstractions exposed by the underlying network or the federated domains depending on their representation of resources.

cs.NI↗