Machine Learning (ML), and specifically deep learning, has become a prominent and rapidly growing research topic within the field of wireless communications both in academia and industry and in consideration for future standards particularly around the AI-Native Air Interface. In a discipline traditionally driven by compact analytic mathematical model-driven methods, ML brings along a methodology that is data-driven and carries a major shift in the way wireless systems are designed and optimized. This brings with it both promise of more accurately representing complexities of the real world, as well as a challenge in providing the same levels of analytic performance guarantee and validation we are used to in communications systems, but also carries the challenge and question of what is the right balance between model-driven, data-driven, and joint model-and-data-driven methods. IEEE technical committees have responded to these developments with several ML workshops in major flagship conferences in recent years such as SAC MLC, IEEE GLOBECOM and IEEE ICC activities. While AI/ML have already applied to many areas in communication systems, these have until recently focused largely on more constrained tasks and environments – and have often tended towards a “knobs and meters” approach to tuning of pre-existing expert/model-centric systems. In more recent years, algorithms, tools, computational power, availability of data, and other enablers have led machine learning to more directly solve for larger tasks and signal processing functions within communications systems directly from a data-driven approach. This mirrors the significant breakthroughs within ML in applications such as computer vision, speech and natural language processing of embracing large datasets, efficient concurrent tensor processing, and end-to-end (E2E) learning techniques for solving high complexity tasks and incorporating unique domain knowledge into learning architectures. An intriguing recent field is the design of E2E solutions, where whole communication system models can be learned and the application of these to massive MIMO and extreme MIMO regimes. Such designs combine ML-enabled transceiver design with data-driven channel and system identification. Deploying full communications systems with these techniques will require significant advances in both distributed-learning and privacy-preserving learning techniques and has many natural extensions to the broader field of ML for communication, control, and security & privacy and key applications. This workshop seeks to provide a first-tier platform at IEEE WCNC 2022 for the dissemination of fundamental and applied research results in the MLC.
Beyond providing a platform for the latest high-quality results in the field of machine learning for communication systems and encouraging fruitful and controversial discussions on the core challenges and prospect of the field, this workshop seeks to follow main themes of promoting and encouraging openness, rigor, explainability and reproducibility and considering the impact of domain specific models and techniques with impact on data-driven communications systems. As data-driven systems and their training and inference processes are often very intricate and specialized, containing numerous details regarding dataset composition, model architecture, hyper-parameters, training and evaluation methodology, processing stages and many other details which can be difficult to completely capture within a concise and compact paper, we encourage the adoption also of open software and data publications which can often more exactly capture these details and make them reproducible. This has become the norm in machine learning centric venues (e.g. NeurIPS, ICML), and rigorous new algorithmic work requires the publication and verification of open research. To embrace this within the IEEE ecosystem, this workshop is focused on directly supporting open-ness within machine learning for communications research, and asking researchers to share datasets, code, implementations, and baselines used throughout their work to help facilitate robust review, reproducibility and quantitative comparison by others within the field who can more easily collaborate and compare work when it is conducted in an open and reproducible manner.
As such, we invite the submission of novel, rigorous machine learning for communications research papers on new and/or improved applications, ideas, and approaches along with the joint publication of datasets and source code required to reproduce the work by others. To mirror this spirit of openness and to help accelerate the research process, IEEE will offer to host large datasets on the IEEE MLC dataset server which allows reproducing, modifying, and extending the work. We invite authors to embrace widely used tools such as GitHub and/or GitLab for hosting their verifiable source code, baselines and implementations, embrace repositories such as ArXiv for early pre-publication feedback of works, and to embrace open source tools such as GNU Radio and iPython notebooks.