Research
Optimizing Supply Chains Through High-Dimensional Statistics
on August 29, 2024
Research
Published on August 29, 2024– Updated on August 29, 2024
5G and Beyond: New Challenges in Mobile Network Optimization
In its 2023 call for proposals, CY Initiative funded 14 innovative research projects. Let's take a look at one of the winning EMERGENCE projects, led by Sara Berri, teacher-researcher at the ETIS laboratory and lecturer at CY Cergy Paris University.
CY Initiative: Could you introduce yourself and tell us more about your career and your main research themes?
Sara Berri: After defending my doctoral thesis in 2018 at CentraleSupelec and the University of Bejaia, I did my two postdoctoral years at Telecom-Paris, where I was lucky enough to represent the "New Technologies & Hybridization" working group, part of the European InDiD project.
I then joined CY Cergy Paris University in 2020 as a lecturer in the Computer Science department and research member of the ETIS laboratory, a joint laboratory between CY Cergy Paris University, ENSEA and CNRS.
My research focuses on several themes. One of the central topics I'm working on is resource allocation in future-generation networks, i.e. 5G and beyond. We're talking about resources in general here, and this includes bandwidth, for example, but also resources in terms of storage, physical resources and so on. This is the subject of my winning CY Initiative project.
I'm also working on security issues, including the preservation of privacy, particularly in the context of geolocation sharing. There are many applications on our phones that use location data to provide a service to the user. Sharing location data can leave users vulnerable to misuse by third-party applications, or to external attacks.
So the question is: how can we continue to make use of these location-based applications while preserving users' privacy?
CY Initiative: You won the CY Initiative's 2023 call for proposals. What does this project involve? And what is its objective?
Sara Berri: When I arrived at CY Cergy Paris University, I started working on network slicing and resource allocation. I realized that there were solutions available on this subject, but that they only took into account certain well-defined categories of services. The idea was to say, in view of future developments, why not take into account and anticipate the services planned for future networks, in order to propose more appropriate and efficient network slicing solutions. The aim was to test future services that have several requirements at the same time, i.e. multiple constraints. However, this makes the mathematical optimization problem more complex.
The challenge of our work and of the project is therefore to take into account all the new services that will be integrated by 5G networks and beyond (B5G, 6G), such as autonomous vehicles. All these services need certain specificities in terms of throughput and latency, and we want to take all these constraints into account and propose optimal solutions so that network resources are well distributed between all the services that are going to need them. The challenge is to propose solutions that can cover hybrid services with multiple needs and constraints, rather than a single constraint. It's important to remember that all this takes place on the same network. Resources are therefore shared, but with users who have different requirements, which is not always easy to manage. The division should be as optimal as possible. Allocating resources is a bit like dividing up a big cake, and each user would like to have a piece of it with specific requirements, rather than randomly.
The tools we use globally are optimization tools (linear programming, non-linear programming, multi-objective optimization). For solving problems, we use, for example, reinforcement learning, or we develop approximate solutions for cases requiring real-time solutions.
CY Initiative : In practical terms, how do you study this allocation of resources?
Sara Berri: In this project, we are working on networks in general, but also on use cases such as vehicular networks. With connected or autonomous vehicles, network conditions change very quickly, due to the speed and frequent changes in position. Solutions need to be much more stable than those that could be proposed for any other service. We're working on the theoretical aspects of mathematical modelling to find out how this architecture could work and how solutions can be proposed to ensure that the network runs smoothly. Ultimately, we may even go so far as to propose the integration of other services, depending on the resources available.
These optimization problems are fairly complex to model. We therefore start with mathematical modeling and the solution will be based on reinforcement learning, assuming that there is an agent trying to learn and generate the best solution according to changes in the environment.
In this model, incoming data are changes in the environment and user usage, and outgoing data are the number of resources allocated to a given service. These are solutions that take into account the objectives set in advance (satisfying as many users as possible, or reaching a certain latency limit that must not be exceeded, for example) and integrate them into the formulation.
In general, we work on general proposals that can be adapted to certain situations, by updating certain parameters and adding the necessary constraints.
CY Initiative: What are the main stages of the project?
Sara Berri: The aim is to go beyond what is currently proposed in the state of the art and to find out what can be done if we start with mixed or hybrid services whose constraints are not easy to manage. And if ever there's something we can propose to improve performance, that would be a great step forward.
We already have some initial results. We've done a network breakdown by proposing an additional service that hasn't been studied in the current state of the art, and we've noticed that we can satisfy more users than when we didn't consider it. In reality, we need these services, but previously we considered them separately, neglecting one or the other, and today we don't want to have to choose. We take more account of what users need, and our solutions are even more tailored and personalized.
For example, if we take the example of the autonomous vehicle, latency and throughput need to be good, and not just one of them. In fact, both are essential and must be at a certain level for the service rendered to be of good quality for the user. You can't prioritize one over the other, and that's what we've been doing until now.
Based on these initial results, we have written a first paper which has been accepted for publication, and we are currently preparing further work.
CY Initiative: Why did you apply for this call for proposals?
Sara Berri: The project was made possible by the EMERGENCE program, which represents a springboard to give impetus to our research, in the hope of making it sustainable in the future. CY Initiative's support has enabled us to move forward and, at the same time, to give us the means to grow the project through national and international collaborations, so that we can build up a slightly larger team than we already have within the laboratory and the university.
Today, we're ready to position ourselves on more ambitious projects, notably European projects such as the 6G SNS JU Projects on energy efficiency in these multiple low-latency services. We submitted this project in April with partners in Europe. This project also integrates the energy constraint, so it's a slightly broader project. We have also submitted another European project for the study of intelligent transport systems (or vehicular networks), which has just been accepted and will be funded. The project involves 89 French and European partners. The aim here is to apply the project developed thanks to EMERGENCE and to take the subject further.
In April, we also submitted an ANR project with a specific AI theme. The aim is to study network aspects through the prism of artificial intelligence tools.
Sara Berri: After defending my doctoral thesis in 2018 at CentraleSupelec and the University of Bejaia, I did my two postdoctoral years at Telecom-Paris, where I was lucky enough to represent the "New Technologies & Hybridization" working group, part of the European InDiD project.
I then joined CY Cergy Paris University in 2020 as a lecturer in the Computer Science department and research member of the ETIS laboratory, a joint laboratory between CY Cergy Paris University, ENSEA and CNRS.
My research focuses on several themes. One of the central topics I'm working on is resource allocation in future-generation networks, i.e. 5G and beyond. We're talking about resources in general here, and this includes bandwidth, for example, but also resources in terms of storage, physical resources and so on. This is the subject of my winning CY Initiative project.
I'm also working on security issues, including the preservation of privacy, particularly in the context of geolocation sharing. There are many applications on our phones that use location data to provide a service to the user. Sharing location data can leave users vulnerable to misuse by third-party applications, or to external attacks.
So the question is: how can we continue to make use of these location-based applications while preserving users' privacy?
CY Initiative: You won the CY Initiative's 2023 call for proposals. What does this project involve? And what is its objective?
Sara Berri: When I arrived at CY Cergy Paris University, I started working on network slicing and resource allocation. I realized that there were solutions available on this subject, but that they only took into account certain well-defined categories of services. The idea was to say, in view of future developments, why not take into account and anticipate the services planned for future networks, in order to propose more appropriate and efficient network slicing solutions. The aim was to test future services that have several requirements at the same time, i.e. multiple constraints. However, this makes the mathematical optimization problem more complex.
The challenge of our work and of the project is therefore to take into account all the new services that will be integrated by 5G networks and beyond (B5G, 6G), such as autonomous vehicles. All these services need certain specificities in terms of throughput and latency, and we want to take all these constraints into account and propose optimal solutions so that network resources are well distributed between all the services that are going to need them. The challenge is to propose solutions that can cover hybrid services with multiple needs and constraints, rather than a single constraint. It's important to remember that all this takes place on the same network. Resources are therefore shared, but with users who have different requirements, which is not always easy to manage. The division should be as optimal as possible. Allocating resources is a bit like dividing up a big cake, and each user would like to have a piece of it with specific requirements, rather than randomly.
The tools we use globally are optimization tools (linear programming, non-linear programming, multi-objective optimization). For solving problems, we use, for example, reinforcement learning, or we develop approximate solutions for cases requiring real-time solutions.
CY Initiative : In practical terms, how do you study this allocation of resources?
Sara Berri: In this project, we are working on networks in general, but also on use cases such as vehicular networks. With connected or autonomous vehicles, network conditions change very quickly, due to the speed and frequent changes in position. Solutions need to be much more stable than those that could be proposed for any other service. We're working on the theoretical aspects of mathematical modelling to find out how this architecture could work and how solutions can be proposed to ensure that the network runs smoothly. Ultimately, we may even go so far as to propose the integration of other services, depending on the resources available.
These optimization problems are fairly complex to model. We therefore start with mathematical modeling and the solution will be based on reinforcement learning, assuming that there is an agent trying to learn and generate the best solution according to changes in the environment.
In this model, incoming data are changes in the environment and user usage, and outgoing data are the number of resources allocated to a given service. These are solutions that take into account the objectives set in advance (satisfying as many users as possible, or reaching a certain latency limit that must not be exceeded, for example) and integrate them into the formulation.
In general, we work on general proposals that can be adapted to certain situations, by updating certain parameters and adding the necessary constraints.
CY Initiative: What are the main stages of the project?
Sara Berri: The aim is to go beyond what is currently proposed in the state of the art and to find out what can be done if we start with mixed or hybrid services whose constraints are not easy to manage. And if ever there's something we can propose to improve performance, that would be a great step forward.
We already have some initial results. We've done a network breakdown by proposing an additional service that hasn't been studied in the current state of the art, and we've noticed that we can satisfy more users than when we didn't consider it. In reality, we need these services, but previously we considered them separately, neglecting one or the other, and today we don't want to have to choose. We take more account of what users need, and our solutions are even more tailored and personalized.
For example, if we take the example of the autonomous vehicle, latency and throughput need to be good, and not just one of them. In fact, both are essential and must be at a certain level for the service rendered to be of good quality for the user. You can't prioritize one over the other, and that's what we've been doing until now.
Based on these initial results, we have written a first paper which has been accepted for publication, and we are currently preparing further work.
CY Initiative: Why did you apply for this call for proposals?
Sara Berri: The project was made possible by the EMERGENCE program, which represents a springboard to give impetus to our research, in the hope of making it sustainable in the future. CY Initiative's support has enabled us to move forward and, at the same time, to give us the means to grow the project through national and international collaborations, so that we can build up a slightly larger team than we already have within the laboratory and the university.
Today, we're ready to position ourselves on more ambitious projects, notably European projects such as the 6G SNS JU Projects on energy efficiency in these multiple low-latency services. We submitted this project in April with partners in Europe. This project also integrates the energy constraint, so it's a slightly broader project. We have also submitted another European project for the study of intelligent transport systems (or vehicular networks), which has just been accepted and will be funded. The project involves 89 French and European partners. The aim here is to apply the project developed thanks to EMERGENCE and to take the subject further.
In April, we also submitted an ANR project with a specific AI theme. The aim is to study network aspects through the prism of artificial intelligence tools.
