2024 LINCS Annual Workshop with its Scientific Committee

2024 LINCS Annual Workshop with its Scientific Committee


LINCS

Overview

On June 27&28 2024, the LINCS launches its Annual Workshop with the Scientific Committee, composed by internationally renowned scholars. It will be held in the LINCS premises in Palaiseau (amphi Rose Dieng, 19 place Marguerite Perey, 91123 Palaiseau)

The workshop will include:

  • LINCS surveys;
  • Scientific Highlights by the LINCS senior members;
  • “Elevator pitch” collective sessions by the LINCS PhD students, followed by parallel poster sessions;
  • Invited talks by the members of the Scientific Committee.

The workshop is a reserved event for the LINCS scientific community, but the video recordings of the talks given by the members of our Scientific Committee will be later made available on the LINCS YouTube channel.

Organization committee

  • Sébastien Tixeuil (Sorbonne Université)
  • Cristina Venitucci (LINCS)

Talks

Workshop-LINCS-agenda-2024

Introduction to PEPR, by Daniel Kofman

Launched on July 10, 2023, the PEPR Future Networks research program was established as part of the France 2030 investment plan to support research activities in the field of future networks integrating 5G and 6G.

The PEPR Future Networks is led by the CEA with Dimitri Ktenas, the CNRS with Serge Verdeyme, and the IMT with Daniel Kofman.

1st Survey: How to model spatial heterogeneity ?, by Laurent Decreusefond (Institut Mines Telecom)

Historically, Poisson point processes have been used to model locations of base stations and mobile units at the district scale, where the distributions are somehow « regular » . With the new wireless systems, we now need to model situations at the scale of the house (for indoor diffusion), at the scale of a street (taking into account buildings, obstacles, etc.) or at the scale of the sky (for non-terrestrial networks). We exhibit several models which can be used for such situations and give some of their mathematical properties.

1st LINCS Scientific Highlight: Misbehaviour detection in V2X communications: safety considerations, by Francesca Bassi (System X)

2nd LINCS Scientific Highlight : Large Language Models Playing Mixed Strategy Nash Equilibrium Games, by Alonso Silva (Nokia Bell Labs)

Generative artificial intelligence (Generative AI), and in particular Large Language Models (LLMs) have gained significant popularity among researchers and industrial communities, paving the way for integrating LLMs in different domains, such as robotics, telecom, and healthcare. In this paper, we study the intersection of game theory and generative artificial intelligence, focusing on the capabilities of LLMs to find the Nash equilibrium in games with a mixed strategy Nash equilibrium and no pure strategy Nash equilibrium (that we denote mixed strategy Nash equilibrium games).

The study reveals a significant enhancement in the performance of LLMs when they are equipped with the possibility to run code and are provided with a specific prompt to incentive them to do so. However, our research also highlights the limitations of LLMs when the randomization strategy of the game is not easy to deduce. It is evident that while LLMs exhibit remarkable proficiency in well-known standard games, their performance dwindles when faced with slight modifications of the same games.

This paper aims to contribute to the growing body of knowledge on the intersection of game theory and generative artificial intelligence while providing valuable insights into LLMs strengths and weaknesses. It also underscores the need for further research to overcome the limitations of LLMs, particularly in dealing with even slightly more complex scenarios, to harness their full potential.

1st Invited talk S.C. : The Multi-server Job Queuing Model, by Marco Ajmone Marsan (PoliTO, IT)

In most cases, queuing models assume one-dimensional services: each customer is served by one server for a random period of time. Today’s services are more complex: for example, jobs that run in cloud data centers need a variable number of CPUs and/or GPUs, as well as a variable amount of memory, and use those resources for a random period of time. The analysis of these contexts calls for queuing models with multi-dimensional services. Research in this area started very recently and results are scarce.

This talk will consider the two-dimensional service case, and report on some recent results.

3rd LINCS Scientific Highlight: Vulnerability detection under poisoning attacks through image processing, by Lorena Gonzalez Manzano (Institut Mines Telecom)

The complexity of current systems encourages the emergence of vulnerabilities. Detectors are developed in this regard, most of them using Artificial Intelligence (AI) techniques. However, AI is not without its problems, especially those attacks affecting the training set. In this talk a novel vulnerability detector, called IVul, based on ML-based image processing, using Convolutional Neural Network (CNN), is presented. IVul is evaluated under the presence of backdoor attacks, applying more than three thousand code samples associated with two representative programming languages (C\# and PHP). Results outperform other comparable state-of-the-art vulnerability detectors and show that the type of attack may affect a particular language more than another.

4th LINCS Scientific Highlight: Inferring object trajectories in a multi-camera environment, by Marc-Oliver Buob (Nokia Bell Labs)

This talk explains how to rebuild the trajectories of mobile objects of interest moving on a flat scene and observed by multiple cameras with overlapping fields of view. More specifically, we aim to find the trajectories of hockey players evolving on the Nokia Arena rink (Tampere, Finland). Using a YOLO-like detector, one can obtain approximate world coordinates of some of the players from each frame produced by each camera. The goal of this work is to rebuild the trajectories of each player given these streams of coordinates. These trajectories can be used to render the players within a 3D model of the arena, for example one that is generated by a NeRF (Neural Radiance Field). After covering the difficulties inherent to our problem, we will explain all the steps and the related optimization models used to transform camera feeds to player trajectories. This is a Joint work by Marc-Olivier Buob, Pierre Escamilla, Jeongran Lee.

2nd Invited talk S.C.: Network Community Structure under Metric Sparsification, by Patrick Thiran (EPFL, CH)

Graph sparsification replaces a graph by a sparser copy, with fewer vertices and/or edges, to reduce storage and/or computational costs while preserving (some) structural properties. We consider here weighted graphs, and sparsify them by keeping the same vertex set but by reducing their edge set to their metric backbone. The metric backbone of a weighted graph is the union of all-pairs shortest paths, and is obtained by removing all edges (u,v) that do not lie on the shortest path between vertices u and v. In networks with well-separated communities, the metric backbone tends to preserve many inter-community edges, because these edges serve as bridges connecting two communities, but tends to delete many intra-community edges because the communities are dense. This suggests that the metric backbone might not preserve the community structure of the network. However, prior empirical work showed instead that the metric backbone of real-world networks preserves the community structure of the original network.

In this work, we analyze the metric backbone of a broad class of weighted random graphs with communities, and we prove the robustness of the community structure with respect to the deletion of all the edges that are not in the metric backbone. An empirical comparison shows that the metric backbone is an effective graph sparsifier in the presence of communities. This is a joint work with Maximilien Dreveton, Charbel Chucri and Matthias Grossglauser.

3rd Invited talk S.C.: Machine learning in telco networks and how to evaluate it, by Holger Karl (Hasso-Plattner-Institute University of Potsdam, DE)

Applying machine learning in telco networks is often considered promising, with distributed deployment of both inference and training considered mandatory to deal with the large scale of such networks. Such deployments challenge current evaluation techniques: How do we know that an ML approach is actually good, when confronted with the challenges of realistic networks? This talk will briefly present two toolkits, one emulation-based, one simulation-based, for such evaluation challenges. Mostly, the talk is intended to kickstart a discussion on what kinds of tools and evaluation setups are needed for realistic, convincing performance evaluation campaigns.

5th LINCS Scientific Highlight: Reading Groups @ Lincs, by Francesca Bassi (SystemX) et François Durand (NBL)

We present the three reading groups of the LINCS: “Network Theory”, “Practical Networks”, and “Tools, Tips and Tricks”.

2nd Survey: Quantum Computing & Networking @ LINCS, Ludovic Noirie (Nokia Bell Labs)

The 1st quantum revolution is about manipulating groups of quantum particles such as photons, electrons and atoms, with their interactions. Technologies from it are mature and commercially available (transistors, lasers, etc.). The 2nd quantum revolution is about manipulating quantum superposition, quantum entanglement and individual particles to get some functionalities not available by classical physics. This includes quantum computing and quantum communications (quantum key distribution systems, quantum networks), for which the technology is not yet mature and there is a lot of research. In this talk, we will describe the past, current and future activities at LINCS related to these prospective research domains.

6th Lincs Scientific Highlight: GeoResolver: An Accurate, Scalable, and Explainable Geolocation Technique Using DNS Redirection, by Hugo Rimlinger (Sorbonne Université)

Obtaining an accurate, explainable, Internet-scale IP geolocation dataset has been a longstanding goal of the research community. Despite decades of research on IP geolocation, no current technique can provide such a dataset. In particular, latency-based geolocation techniques do not scale, because, on one hand, we have thousands of available vantage points to perform measurements, but on the other hand, we have no way to select the right ones for each IP address.

In this paper, we present GeoResolver, which bridges this gap by using the idea that when multiple operators redirect two prefixes to the same servers, these prefixes should be close to each other. With this intuition, we define a methodology to measure and compare the redirection of prefixes to servers using ECS DNS measurements, and select the prefixes with the smallest redirection distance to a target prefix to issue the latency measurements to targets in that prefix. GeoResolver performs nearly as well as a brute force approach using all the vantage points, geolocating 94% of the targets that could actually be geolocated at metro level by using all the vantage points, while using 4.3% of the probing bud- get compared to a state-of-the-art technique. We carefully design GeoResolver to be both accurate and scalable and with our implementation, we start a large-scale geolocation campaign.

7th Lincs Scientific Highlight: Can Aloha Achieve Implicit Admission Control?, by Fabien Mathieu (Inria)

Adaptive Aloha protocols allow several transmitting stations to share the same communication channel, with each transmitter managing its transmissions using an internal state. They form the basis of many networks, including 802.11 networks. In this presentation, we consider a discrete-time scenario where “saturated” stations always have to transmit. Using a geometric decay model, we give a new stability condition based on the notion of ambient noise and propose formulas for approximating the behavior.

However, comparing these results with simulations shows that for large the stable/instable distinction loses its meaning, as only a transient regime is visible at reasonable time scales. In this transient regime, we observe a form of implicit admission control: if it is too large, some of the stations are in “frozen” states, allowing the others a reasonable access to the channel.

8th Lincs Scientific Highlight: GNN-based Fast Power Allocation in Cell Free Massive MIMO networks, by Shashwat Mishra (Nokia Bell Labs)

Beyond 5G wireless technology Cell-Free Massive MIMO (CFmMIMO) down-link relies on carefully designed precoders and power control to attain uniformly high rate coverage. Many such power control problems can be calculated via second order cone programming (SOCP). In practice, several order of magnitude faster numerical procedure is required because power control has to be rapidly updated to adapt to changing channel conditions. We propose a Graph Neural Network (GNN) based solution to replace SOCP. We construct a graph representation of the problem that properly captures the dominant dependence relationship between access points (APs) and user equipment (UEs). Simulation results show the superiority of our approach in terms of computational complexity, scalability and generalizability for different system sizes and deployment scenarios.

4th Invited talk S.C.: A brief (machine learning) foray at the edge of computing, by Roch Guerin (Washington University in Saint Louis)

Edge computing solutions have proliferated, fueled by a combination of increased network ubiquity, advances in computing, especially in embedded devices, and by the growing need to bring computations closer to where data is produced. Many of those scenarios are driven by machine learning applications. In this talk, I will discuss two projects, both motivated by edge computing machine learning applications, and for which machine learning was itself instrumental in devising an efficient solution.

  1. The first project [1] targets object detection with local and edge compute resources cooperating to optimize detection accuracy under load constraint on the edge server. Under such a constraint, the goal is to devise a simple policy to decide which images to offload to the edge server while maximizing detection accuracy. This calls for a metric that quantifies improvements in overall detection accuracy from offloading an individual image, and an estimator for that metric that can run on embedded devices. The benefits of the approach are again demonstrated experimentally.
  2. The second project deals with an object classification problem where a camera is uploading images to an edge server for classification [2]. The wireless network used to upload successive images is, however, subject to bandwidth fluctuations. This requires an adaptive transmission strategy to maximize inference accuracy, irrespective of the amount of data that can be transmitted for each image. We realize this through a simple application of stochastic tail-drop when training a neural compression algorithm and demonstrate the efficacy of the approach on a local testbed.

    3. [1] J. Qiu, R. Wang, B. Hu, R. Guerin, and C. Lu, “Optimizing Edge Offloading Decisions for Object Detection.” Under submission, 2024.

    [2] R. Wang, H. Liu, J. Qiu, M. Xu, R. Guerin, and C. Lu, “Progressive Neural Compression for Adaptive Image Offloading Under Timing Constraints.” Best Student Paper Award, 2023 IEEE Real-Time Systems Symposium (RTSS), December 2023, Taipei, Taiwan.

    3rd Survey: LINCS activities in Cellular Networks, by François Baccelli (Inria)

    This presentation will survey recent advances of LINCS researchers on cellular networks.

    It will cover results on the system level analysis of:

    • RIS deployments in cellular and NTN networks;
    • Low Earth Orbit networks;
    • Latency guarantees in D2D and cellular networks (in progress).

        • It will also briefly survey ongoing work on physical Layer and MAC.

        5th Invited talk S.C.: Adaptive Beam Management for Mobile mm-Wave Devices, by P.R. Kumar (Texas A&M University)

        In the mm-wave band, narrow beams are used to overcome path loss. They can be used to support high data rates with reduced interference. However, it is necessary to manage the narrow directional beams since they can become misaligned as users move, blocked as users rotate, or lose connectivity as users move out of range.

        We will present an account of three protocols for these problems – BeamSurfer, Unblock and Terra. We will also address the general problem of detecting a Man-In-the-Middle in wireless communication, and present a protocol, Reveal, to do so.

        [Joint work with Santosh Ganji].