Tutorial Statistical Learning with Generative Models for Communications

I am happy to deliver a tutorial on Statistical Learning with Generative Models for Communications at IEEE ICC 2023 on June 1st, 2023.
Machine learning is gaining more and more interest in several application domains. Communications is one of those, although herein it is not fully clear whether ML can bring disruptive innovation and offer improved performance w.r.t. well established model-based approaches.
This tutorial at ICC 2023 has three distinctive elements: firstly, it focuses on a timely topic, i.e., generative models for signal analysis (learning) and synthesis (generation); secondly, it highlights the pivotal role played by the estimation of the channel input/output mutual information; thirdly, it considers several communication “problems” and it illustrates solutions with generative models obtained from the formulation of optimality criterions for:

  • Synthetic channel and noise modeling
  • Coding/decoding design in unknown channels
  • Channel capacity estimation.

In the above-mentioned problems, a key enabling component is the ability to estimate mutual information. This will lead us to the description of known and novel mutual information estimators. Their application will be considered to derive optimal decoding strategies with deep learning neural architectures obtained from an explainable mathematical formulation. Then, the joint design of the coding and decoding scheme aiming to achieve channel capacity will be considered. Finally, the most ambitious goal of estimating the capacity in unknown channels. This last goal rendered possible by the exploitation of cooperative methods that learn the capacity using neural mutual information estimation.

Recommended Posts