--
Diffusé avec l'autorisation de l'auteur d'une copie de contrôle montrée dans la démonstration

Retour d'expérience sur l'évaluation par les pairs dans Moodle, sur les projets informatiques d'application du module CSC4101 "Architectures et applications Web" à TSP

Olivier Berger

Seminar talk by Salvatore D'Oro (Northeastern University - Boston) (30/11/2022)

"Network Intelligence in the Open RAN: architectures, challenges and open experimental platforms"

Abstract
The Open Radio Access Network (RAN) paradigm and its embodiment based on the O-RAN Alliance specifications is poised to transform the telecom ecosystem via virtualized and disaggregated RAN, open interfaces and RAN intelligent controllers (RICs). This talk will provide a comprehensive introduction to O-RAN, including a clear overview of the potentials of the Open RAN paradigm and of the challenges to its realization. After a brief introduction on O-RAN and its architecture, we will focus our attention on discussing how Artificial Intelligence (AI) and data-driven optimization will transform the ecosystem and will identify challenges to be overcome to de-risk AI, as well as how to design, train and deploy AI-based solutions that are able to generalize across multiple topologies, network and traffic conditions. Finally, the talk with focus on showcasing selected results obtained using OpenRAN Gym (https://openrangym.com/), an experimental platform for research on O-RAN developed by our team at Northeastern University.

Seminar talk by Qiong Liu (TP) (16/11/2022)

"Coverage and Stability Analysis of Cellular Network with Temporal Traffic"

Abstract
In this talk, we discuss the stability region characterization of a random network when a traffic model is integrated into the network geometry description.

First, we characterized the stable coverage probability of a random network. Starting from the notion of dynamic coverage probability, the interaction between the queue states in the network is taken into account using discrete Markov chain modelling of the queues, where the typical user’s service rate depends on the dynamic coverage probability.

A more fine-grated description of the phenomenon is made by answering the question, “what is the proportion of unstable queues in the network?” In this case, the notion of epsilon-stability is exploited, which describes the set of traffic intensities for which a queue taken at random has a probability of diverging less than epsilon.

Finally, the characterization of the stable region by considering the resource allocation is very difficult to obtain, because of the dependence between the geometry and the network’s dynamicists and the allocation strategy. However, the dynamic nature of the network considered lends itself perfectly to description by a Markovian decision process, for which reinforcement learning technologies can be proposed. We therefore study the stability-cost characteristic of the network operation according to the transmission policy.

Seminar talk by Matthew Andrews (Nokia US) (09/11/2022)

"Talking to robots and drones: Network requirements for autonomous agents in factories and warehouses of the future"

Abstract
Future factories and warehouses will contain fleets of autonomous agents to assist with monitoring and control. In this talk we will describe a variety of machine learning problems that arise in such a setting. These include robot navigation, localization and multi-agent coordination, as well as the construction of an analytics pipeline to understand the images streamed from an indoor drone. A common thread in all these problems is the utilization of cloud robotics, where most of the intelligence is migrated from the agents themselves to an edge cloud. In this environment the quality of the network is critical for effective operation. We will discuss the network performance required to support such systems, and also describe some concrete deployments that support indoor farming and warehouse inventory management.

Seminar talk by Michel Barbeau (Carleton University, Ottawa) (26/10/2022)

"Work Memory Requirements in Error Susceptible Quantum Networks"

Abstract
We consider the problem of path congestion avoidance in networks of quantum repeaters and terminals. In other words, the avoidance of situations when demands exceed capacity. We assume networks in which the sets of complete paths between terminals may affect the capacity of repeaters in the network. We compare the reduction of congestion avoidance of path establishment algorithms: shortest-path establishment vs. layer-peeling path establishment. We observe that both strategies provide an equivalent entanglement rate, while the layer-peeling establishment algorithm considerably reduces the congestion in the network of repeaters. Repeaters in the inner layers get less congested and require a lower number of qubits while providing a similar entanglement rate.

Michel Barbeau,
Professor and Director
School of Computer Science
Carleton University, Ottawa, Canada

Seminar talk by Stefano Paris (Nokia) (12/10/2022)

"Extended Reality in 5G NR and Beyond"

Abstract
Extended Reality (XR) has been one of the ambitious research topics under development for several decades and today it is becoming mature enough to appear in the mass market. The replacement of smartphones with smaller wearable devices and headset mounted devices enables numerous novel use cases in consumer, engineering, industrial, and healthcare areas, as well as represents a foundation for upcoming Metaverse revolution. The successful adoption of XR requires support from the underlying 5G system, since XR introduce challenging requirements for radio mobile systems in terms of capacity and latency that do not perfectly fit into the existing classification of fifth-generation (5G) applications and services. Additionally, power saving needs of user devices result in a clear trade-off between capacity and energy consumption. In this talk, we will revise the main challenges discussed in 3GPP and possible solutions to fully support XR applications over 5G NR. In particular, we will highlight the trade-off between power-saving and capacity focusing on possible enhancements to power saving features.

Seminar talk by Rémi Varloot (Nokia) (28/09/2022)

"PCANT — Programable Capture and Analysis of Network Traffic"

Abstract
PCANT is a solution for streamlining the deployment of arbitrary online traffic analysis methods inside networks. It allows network administrators to dynamically instantiate new monitoring techniques on their network without the need for any manual setup, and data scientists to test new algorithms on a network without requiring any skills pertaining to network equipment administration and configuration.