Gastvorlesungen im Wintersemester 2019/2020
Im Rahmen der Vorlesung Channel Coding II werden in diesem Wintersemester 2 Gastvorlesungen stattfinden. Die exakten Termine werden noch bekannt gegeben.
Gastvorlesung 1: "Deep Learning Meets Channel Coding"
Datum und Ort: 22. Januar 2019, 09:45 in room 20.30 SR -1.025 (UG)
Dozent: Sebastian Cammerer, M.Sc., Universität Stuttgart
Artificial intelligence and machine learning are currently considered as the most important universal technology of our era, similar to electricity and the combustion engine; their applications now extend into almost every industry and research domain. Although researchers have tried to address communications-related problems with machine learning in the past, it did not have a fundamental impact on the way we design and implement communications systems today. At first glance, machine learning techniques do not appear to be a good match to physical layer problems, with 50 years of tremendous progress based on “classic” signal processing, communication and information theory, approaching close-to-optimal Shannon limit performance on many channels. However, when having a closer look it turns out that the graph representation of “modern” error correction codes can be interpreted as a neural network and, thus, allows to adapt existing deep learning techniques for the channel coding purpose – and, thereby, also insights from 50 years of “classical” channel coding may contribute to the deep learning domain expertise.
The goal of this lecture is to provide an overview of the field of deep learning for channel coding starting with the naïve idea of replacing the whole decoder by trainable neural networks. As we will see, this data-driven design of new decoding algorithms is limited by exponential training complexity, namely the curse of dimensionality, and does not scale well towards practical code lengths. Thus, hierarchical net structures in combination with deterministic layers turn out to be much more efficient as will be shown by the example of trainable neural network polar decoders.
In the second part of the lecture, we aim to reuse and augment existing decoding algorithms. Therefore, we introduce the concept of “neural belief propagation” decoding by an iterative loop unrolling of the classical BP decoding algorithm and reinterpret the graph as a trainable neural network.
If time allows, we will see that also the encoding can benefit from deep learning techniques if the code design is “learned” by neural networks. In particular for ultra-short length code design, having the actual channel and the actual decoder “in-the-loop” allows to improve existing code design which typically relies on canonical channel models (such as the AWGN) and asymptotic properties.
From a more practical perspective, it turns out that the implementation of existing algorithms in state-of-the-art machine-learning software libraries is often demanding and time consuming. In the last part of this lecture, we aim to lower the barrier-to-entry for ML-newcomers to enable the implementation of own applications. Therefore, the attendees receive Jupyter notebooks containing code examples, which allows them to quickly get up to speed with this new and exciting field.
Biography of Guest Lecturer
Sebastian Cammerer is a member of research staff at Institute of Telecommunications, University of Stuttgart, and is pursuing his Ph.D. He received the B.Sc. and M.Sc. degree (with distinction) in electrical engineering and information technology from University of Stuttgart, Germany, in 2013 and 2015, respectively. During his years of study, he worked as a research assistant at multiple institutes of University of Stuttgart. His research topics are channel coding and machine learning for communications. Further research interests are in the areas of iterative receiver algorithms, parallelized computing for signal processing and information theory. He won the third prize at the 2019 Bell Labs Prize and is recipient of the IEEE SPS Young Best Paper Award 2019, the University of Stuttgart Best Paper Award 2019, the Anton- und Klara Röser Preis 2016, the Rohde&Schwarz Best Bachelor Award 2015 and the VDE-Preis 2016 for his master thesis.
Gastvorlesung 2: "Channel Coding in the Short Blocklength Regime"
Datum und Ort: 5. Februar 2020 (Achtung, Terminänderung!), 09:45 in Raum 20.30 SR -1.025 (UG)
Dozent: Gianluigi Liva, PhD., German Aerospace Center (DLR)
The impetuous rise of IoT systems is pushing the development of efficient communication strategies involving the (bursty) transmission of small amounts of data.
In particular, powerful error correcting codes protecting packets composed by a few hundred bits are fundamental building blocks to enable reliable energy-efficient communication among IoT nodes. In this lecture, we will address the main challenges involved in the design of short blocklength channel codes. A brief introduction to the information-theoretic limits of finite blocklength communications will be provided, and a number of recent competing coding solutions will be benchmarked against the finite blocklength performance bounds.
Biography of Guest Lecturer
Gianluigi Liva received the M.S. and Ph.D. degrees in electrical engineering from the University of Bologna, Bologna, Italy, in 2002 and 2006, respectively. Since 2003, he has been investigating channel codes for high-data rate Consultative Committee for Space Data Systems (CCSDS) missions. From 2004 to 2005, he was involved in research with the University of Arizona, Tucson, where he was designing low-complexity error correcting codes for space communications. Since 2006, he has been with the Institute of Communications and Navigation, German Aerospace Center (DLR), where he currently leads the Information Transmission Group.