As a part of its training events, the ITN WindMill Project invites scholars and experts in the field of machine learning and wireless communications to hold lectures for the ESRs. The abstracts of these are available on this page, under the tab of the corresponding training event.
In this lecture, we will provide an overview of the URLLC-related 3GPP standardization in 5G releases 15, 16 and 17. We will discuss the URLLC use-cases and their requirements, as well as the proposed enhancements of mobile cellular standards addressing these requirements. In closing, we will discuss the potential applications of ML techniques for enabling URLLC connectivity in 3GPP approaches.
The next frontier towards truly ubiquitous connectivity is the use of Low Earth Orbit (LEO) small-satellite constellations to support 5G and Beyond-5G (B5G) networks. LEO constellations can support ultra-reliable communications (URC) with relaxed latency requirements of a few tens of milliseconds. However, small-satellite impairments and the use of low orbits pose major challenges to the design and performance of these networks. This session provides a comprehensive overview of communications with LEO constellations, and insights of the role of ML to optimize the resource utilization, performance and maintenance of the network.
This presentation gives an introduction to the principles of Multi-Connectivity both from a 3GPP perspective and from a generic perspective. Furthermore, a selection of recent work will be presented covering both classical as well as novel use cases such as UAVs.
URLLC use cases pose significant challenges in the wireless access. Their stringent latency requirements hinder the use of traditional grant-based access schemes and limit the utility of feedback mechanisms. At the same time, the extremely high reliability requirements are oftentimes guaranteed through naive resource over-provisioning. In this session, we explore potential solutions based on grant-free random access protocols and non-orthogonal multiple access (NOMA) for resource-efficient URLLC communications.
The application of supervised learning techniques for the design of the physical layer of a communication link is often impaired by the limited amount of pilot data available for each device; while the use of unsupervised learning is typically limited by the need to carry out a large number of training iterations. In this talk, meta-learning, or learning-to-learn, is proposed as a tool to alleviate these problems.
First, the talk will consider an Internet-of-Things (IoT) scenario in which devices transmit sporadically using short packets with few pilot symbols over a fading channel. The number of pilots is generally insufficient to obtain an accurate estimate of the end-to-end channel, which includes the effects of fading and of the transmission-side distortion. To tackle this problem, pilots from previous IoT transmissions are used as meta-training data in order to train a demodulator that is able to quickly adapt to new end-to-end channel conditions from few pilots. Various state-of-the-art meta-learning schemes are adapted to the problem at hand and evaluated, including MAML, FOMAML, REPTILE, and CAVIA. Both offline and online solutions are developed.
Then, the unsupervised training of an auto-encoder consisting of the cascade of encoder, channel, and decoder is studied. An important limitation of the conventional approach is that training should be generally carried out from scratch for each new channel. To cope with this problem, prior works considered joint training over multiple channels with the aim of finding a single pair of encoder and decoder that works well on a class of channels. In this talk, the limitations of joint training are obviated via meta-learning: Rather than training a common model for all channels, meta-learning finds a common initialization vector that enables fast training on any channel. The approach is validated via numerical results, demonstrating significant training speed-ups, with effective encoders and decoders obtained with as little as one iteration of Stochastic Gradient Descent.
The availability of cheap sensing devices and of capacious storing solutions have lead researchers to build gigantic data sets. However, the more is not always the better: finding our way through such a great amount of information may be tough from the point of view of practical algorithms, due to memory and computational requirements. Moreover, interpretability may also become an issue, since intuition is hardly effective when dealing with thousands of parameters. This talk is a brief introduction to convex optimization techniques that promote a sparse representation of large-dimensional problems, meaning that they favor parsimonious solutions that only involve a considerably reduced subset of inputs.
Recently, deep learning has received significant attention as a technique to design and optimize wireless communication systems and networks. The usual approach to use deep learning consists of acquiring large amount of empirical data about the system behavior and employ it for performance optimization (data-driven approach). However, the application of deep learning to communication networks design and optimization offers more possibilities. As opposed to other fields of science, such as image classification and speech recognition, mathematical models for communication networks optimization are very often available, even though they may be simplified and inaccurate.
This a priori expert knowledge, which has been acquired over decades of intense research, cannot be dismissed and ignored. In this tutorial, in particular, a new approach is put forth, which capitalizes on the availability of (possibly simplified or inaccurate) theoretical models, in order to reduce the amount of empirical data to use and the complexity of training artificial neural networks. We provide several recent examples to show that synergistically combining prior expert knowledge based on analytical models and data-driven methods constitutes a suitable approach towards the design and optimization of communication systems and networks with the aid of deep learning based on ANNs.
This lecture will provide the attendees with the theoretical background on recent Reinforcement Learning (RL) techniques. These range from well stablished tabular techniques like Q-learning and Reinforce to Deep Learning architectures for Reinforcement Learning such as DQN or A3C. Finally, an overview of multi-agent RL will be presented and the latest techniques on the problem of Learning to Communicate will be described.
The workshop aims at preparing fellows for their PhD experience by helping them envision their personal and professional development, and by encouraging them to be proactive with regards to their career goals. First information on the job market for PhDs will be given in order to increase their awareness of alternative career options.
This workshop is intended as a forum to present and discuss the Reinforcement Learning solutions proposed by the participants to a well-known problem in wireless communications. The chosen problem is time-frequency resource allocation. The solutions proposed by the participants will be scored against each other and compared against several baseline non-RL solutions provided as a reference. The objective of this workshop is to reach a consensus of the characteristics that the optimum RL solution should have, how computationally expensive it would be, and how much of an advantage it would provide against well-established non-RL approaches.
The wireless connectivity is considered to be one of the main enabling technologies for Industry 4.0 as it can provide the required mobility, flexibility and efficiency for the manufacturing world. But it is also one of the most challenging IoT areas for wireless connectivity.
In this tutorial we will go through the main communication aspects relevant for factory automation, incl. communication principles, applications, requirements, Industrial Ethernets, Safety, Time-Sensitive Networks, wireless technologies for factories, private 5G networks for factories, and standardization roadmap.
One of the major project milestones is the Knowledge Sharing platform (KSP). It is envisioned as an ‘Interactive Online Infrastructure through which the newly created knowledge will be allowed to circulate within and outside the network’ (The ‘network’ being all project members). The KSP will be established as a website and is to be managed by the ESRs who will cater for all of its content. In this session we will revisit the purpose, the possibilities, and discuss further how to implement this together as a project within the project.