- Program -
Location
The training school will be held at IMT Nord Europe, on the Villeneuve d'Ascq campus, France. The lectures will be given in Amphythéatre Morse.
Lunch and coffee breaks
Lunch and coffee breaks will be served in the hall of IMT Nord Europe.
Temporary Program
Eduard Jorswieck, Bile Peng, TU Braunschweig, Problem-specific unsupervised machine learning powered by domain knowledge in communication, 3h
As wireless communication systems become increasingly complex, optimizing their performance has become a significant challenge. In this context, unsupervised machine learning is gaining importance due to its ability to autonomously optimize system performance without requiring labeled data. This tutorial introduces a framework for using unsupervised learning in system optimization. In the first part, we introduce the model-based learning with known system models. We focus on techniques to address nonconvex problems with multiple local optima, combinatorial problems with discrete optimization variables, specialized neural network architectures tailored to problem-specific properties, and strategies to combine analytical methods with machine learning. These approaches collectively offer powerful tools for managing the growing intricacies of wireless communication systems.
In the second part of the tutorial, we focus on the model-free learning techniques. This is motivated by the complexity of 6G wireless networks grows due to the large number of network parameters, increasing number of users and services as well as heterogeneous network nodes. Driving by software defined networking (SDN) and network function virtualization (NFV), the configuration of access networks as well as backhaul networks is flexible and offers great potential for age-of-information (AoI) aware and energy-efficient (EE) design. However, the computational complexity to find optimal solutions is too high. Therefore, deep reinforcement learning algorithms are proposed. The second part of the tutorial illustrates two successful applications of DRL in 6G networks: First, we consider a Zero-Touch, DRL-based Proactive Failure Recovery framework for stateful network function virtualization (NFV)-enabled networks. Second, novel joint intelligent trajectory design and resource allocation algorithms based on users’ mobility and their requested services for unmanned aerial vehicles (UAVs) assisted networks are explained, where UAVs act as nodes of a NFV enabled network.
Jakob Hoydis, NVIDIA, Digital Twins for Communications: How to Create and Use Them, 3h
A possible vision for 6G networks is that they can autonomously specialize to the radio environment in which they are deployed. I will discuss two key tools that are required to make this happen, namely differentiable ray racing for the creation of digital twin networks and machine learning. Differentiable ray tracing allows for gradient based optimization of many scene parameters and enables data-driven calibration of ray tracing models to measurements. Such digital twins can then be used as “gyms” for training of environment-specific communication schemes and applications. As examples, I will show how one can learn radio material parameters from channel measurements and present the architecture and performance of a recently developed 5G-compliant neural receiver which is not only compatible with different bandwidth allocations and number of layers but can run in real-time on a GPU.
Mérouane Debbah, Khalifa University, TelecomGPT: Next Generation AI powered Network, 1h30 (virtual)
The evolution of generative artificial intelligence (GenAI) constitutes a turning point in reshaping the future of technology in different aspects. Wireless networks in particular, with the blooming of self-evolving networks, represent a rich field for exploiting GenAI and reaping several benefits that can fundamentally change the way how wireless networks are designed and operated nowadays. To be specific, large language models (LLMs), a subfield of GenAI, are envisioned to open up a new era of autonomous wireless networks, in which a multimodal large model trained over various Telecom data, can be fine-tuned to perform several downstream tasks, eliminating the need for dedicated AI models for each task and paving the way for the realization of artificial general intelligence (AGI)-empowered wireless networks. In this talk, we aim to unfold the opportunities that can be reaped from integrating LLMs into the Telecom domain. In particular, we aim to put a forward-looking vision on a new realm of possibilities and applications of LLMs in future wireless networks, defining directions for designing, training, testing, and deploying Telecom LLMs, and reveal insights on the associated theoretical and practical challenges.
Afef Feki, Nokia, AI feeding Telecom Industry, 1h30
AI/ML is progressively becoming a core component of Radio Access Networks (RAN), evolving from high-level optimization tools to being embedded in fundamental network processes. In line with this progression, the Third Generation Partnership Project (3GPP) officially introduced AI/ML technologies into RAN starting with Release 18, the first release of 5G-Advanced. This integration aims to boost system performance, addressing the rapid growth of massive Internet of Things (IoT) devices and the immense volumes of data they generate.
This talk offers a comprehensive overview of AI/ML integration within the 3GPP RAN framework, focusing on the functional architecture for AI/ML in the New Radio (NR) Air Interface and the various enablers across key 3GPP working groups, such as RAN1 and RAN2. Additionally, it delves into critical aspects of AI/ML life cycle management (LCM) and discusses use cases such as Channel State Information (CSI) enhancements and beam management.
Deniz Gündüz, Imperial College, Learn to Communicate - Communicate to Learn, 3h
Machine learning and communications are intrinsically connected. In the first part of the talk, I will focus on using machine learning techniques for various communication problems. I will start with the fundamental joint source-channel coding problem, and show how neural networks can be used to design semantic communication schemes over various channels, providing adaptation to channel variations, bandwidth, and even number of antennas. I will then move on to the channel coding problem, and present some results on neural network based channel code designs. This will include novel transformer-aided code design for the feedback channel, and the idea of “friendly attacks” to increase the reliability of existing codes. If time permits, I will also talk about how wireless channels can be treated as a function, resulting in a state-of-the-art channel state feedback scheme.
In the second part of the talk, I will start by presenting a new information theoretic tool, called ‘channel simulation’, and show how that can help us design distributed learning algorithms with state-of-the-art performance in terms of communication efficiency and privacy. I will also introduce other information theoretic ideas that can prove useful for distributed learning, in particular, over-the-air computation and unsourced random access.
Alvaro Valcarce, Nokia, Radio Protocol Emergence: Learning the language of the machines, 1h30
Today, RAN protocols are agreed at the industry level in a waterfall process that includes research, design, standardization, implementation and testing. This takes years and consumes a formidable amount of resources. Standardization is important to establish consensus between hardware manufacturers and guarantee cross-vendor inter-operability, but these negotiations not always technically driven. Can this approach be accelerated, optimized and, above all, automated? In this talk, I’ll postulate protocols as the language that machines speak and will describe an ML-based framework to teach independent radio nodes how to develop their own language in a performance-maximization manner.
Leonardo Linguaglossa, Télécom Paris, Machine learning for high-speed networks: challenges and opportunities, 3h
Recent years witnessed a trend of "softwarization" of network components. Instead of static, expensive hardware, operators have started to adopt a more flexible approach based on Virtual Network Functions. This paradigm (aka Network Function Virtualization) advocates implementing network middleboxes such as firewalls or NATs as pieces of software to be deployed and executed on commercial off-the-shelf (COTS) hardware. This has boosted the development of several packet processing frameworks and software switches, which show nowadays multi 10-Gbps capabilities in COTS servers. In parallel, network systems are increasingly adopting machine learning (ML) techniques to solve complex networking tasks such as traffic classification or resource allocation.
As ML techniques require a large amount of data to be collected for both training and validation, when done in software, such measurements can highly affect the measured values, thus biasing the collected data. The intensity of this becomes stronger when measurements are taken close to the data path. Second, even after the training phase, complex model calculations may require dedicated hardware such as external GPUs or custom hardware designed for neural network processing such as TPUs or VPUs.
In this talk, we present an overview of ML challenges applied to networking systems, especially high-speed software networks. We then present a novel approach based on non-invasive data collection relying on pure software.
Philippe Mary, INSA Rennes, Reinforcement learning in wireless communications, 1h30
In this course, we will introduce the attendees to reinforcement learning algorithms and their application to wireless communications. Reinforcement learning can be rooted back to the behavioral psychology and dynamical programming in computer science. The foundation of reinforcement learning will be first reviewed and the class of problems suitable for being solved with a reinforcement learning algorithm introduced. Once the problem can be modeled with a Markov decision process, then reinforcement learning approach can be applied. We will briefly exposed the optimal solution of a Markov decision process and practical algorithms that can be used. Two applications in wireless communications will be given. The first one involves a mutli-armed bandit algorithm for opportunistic radio communications and the second a Q-learning approach to switch on or off base stations in a wireless network in order to save energy. The end of the talk will open perspectives on this fast-growing topic with the current theoretical and practical issues in this domain.
Viktor Arvidsson, Ericsson, Panel
Emmanuel Dotaro, Thalès, Panel
Armen Aghasaryan, Nokia, Panel
Xavier Lagrange, IMT Atlantique, Panel
Laurent Clavier, IMT Nord Europe, Panel
Detailed Temporary Program
