19 - 20 October 2022 - 9:00 am to 12:30 pm EDT
Registration closes on 17 October 2022 at 5:00pm EDT
Online via WebEx
Price$279 IEEE ComSoc member |
Not a member? Join ComSoc Today IEEE Members can add ComSoc to their membership. |
CEU Credits: 0.6 CEUs
Course Description
Artificial intelligence (AI) and machine learning (ML) are considered as some of the most important universal technologies of our era, like electricity and the combustion engine. Particularly deep learning has led to many recent breakthroughs in various domains, such as computer vision, natural language processing, and speech recognition. It is therefore natural to ask what role AI/ML will play for wireless communications? This course tries to shed some light on this question by introducing key concepts of deep learning and their applications to problems in communications, ranging from channel estimation, over a full neural OFDM receiver, to an entirely neural network-based communications system that does not use any traditional algorithm. For now, deep learning for communications is a novel field that offers many attractive interdisciplinary research questions at the interface between machine learning, communications engineering, information theory, as well as hardware design.
As one of the hottest topics currently in our field, it is expected that machine learning will play an increasingly important role in the future evolutions of 5G as well as the development of 6G. Hence, this course is a great opportunity to learn about the cutting-edge research in communications and deep learning with a particular focus on practical implementation with state-of-the-art deep learning libraries.
Developed and taught by one of the pioneers of this field, participants will be guided to relevant resources and provided with a realistic assessment of the impact of machine learning on the future of our industry. Participants will get access to Jupyter notebooks with code examples that will enable them to deepen their understanding of the topic and design their own experiments.
Who Should Attend
The course is intended for engineers and researchers working in the telecommunications industry. It is also of interest to graduate students, postdocs, and professors who want to get an introduction to the topic of deep learning for the physical layer.
Level of Instruction: Introductory/Intermediate
Prerequisites
This introductory/intermediate course on deep learning for the physical layer has the following prerequisites:
- Solid background in digital communication systems, especially the physical layer (OFDM, MIMO, modulation, detection, estimation, channel coding)
- Background on basic information theory, signal processing, and wireless communications.
- Basic knowledge of machine learning and, particularly, deep learning is good to have but not a prerequisite.
- Knowledge of the Python programming language as well as TensorFlow or PyTorch are beneficial
Instructor
Learning Objectives
After this course, participants will be able to:
- Identify when it makes sense to use machine learning to solve a problem.
- Be aware of the current 3GPP standardization efforts with respect to AI/ML.
- Recognize the most important state-of-the-art neural network architectures relevant
- to the physical layer.
- Setup simple experiments to solve physical layer problems using neural
- networks.
- Understand the idea of end-to-end learning and related challenges.
- Describe the idea of graph neural networks and self-attention.
Course Content
- Part I - Introduction & Primer on Machine Learning and Deep Learning
- Introduction
- What is ML / AI?
When (not) to use ML?
What can ML do for communications?
3GPP Rel. 18 ML/AI activities
Components of an ML System
- What is ML / AI?
- Deep Learning Basics
- Feed-forward neural networks
Activation functions
Gradient descent & Backpropagation
Universal approximation theorem
Capacity, over- & underfitting
Regularization
- Feed-forward neural networks
- Convolutional Neural Networks
- Convolutional layers
Stride, pooling
Receptive field and dilation
Depth-wise separable convolutions
Fully convolutional networks & ResNets
Example: Modulation recognition
- Convolutional layers
- Introduction
- Part II - Applications of Deep Learning for Communications
- Introduction to TensorFlow & Sionna
- DL at the Receiver
- OFDM over doubly-dispersive channels
Example: Neural OFDM Receiver
- OFDM over doubly-dispersive channels
- End-to-end Learning
- What is E2E learning?
Bit-Metric Decoding Rate
Geometric shaping & Bit Labelling
Training without a channel model
Example: Train your first Autoencoder
Turbo Autoencoders / Waveform learning
- What is E2E learning?
- Overview of Advanced Topics
- Graph neural networks
- Example: Learned BP Decoder
- Summary & Outlook
- Graph neural networks
Course Materials
Each registered participant receives a copy of instructor slides and access to the recording of the course for 20 business days after the live lecture. Earn 0.6 IEEE Continuing Education Units for participating.
Upon registration, you will automatically be emailed the WebEx invitation for the course session, but you will also be sent a reminder message to join the WebEx session prior to the start of the course. Course materials will be emailed to you and will be available for download from the WebEx session page for this course, the day prior to the scheduled course date.
Course Cancellation and Refund Policy: Requests for online course cancellations must be received 3 business days prior to the course date for a full refund. Once course materials have been shared with a participant, a cancellation request cannot be accommodated.
Contact Us
For general inquiries and technical support, contact Tara McNally, Certification and Professional Education Manager.
Review the system requirements for WebEx.
Test your browser by joining a meeting.