Information theoretic perspectives via rate-distortion trade-offs in goal-oriented semantic communications

In this technical talk, we discuss information theoretic viewpoints of goal-oriented semantic communications, a topic which currently receives much interest due to its promising framework in future communication technologies. Particularly, herein westress the usefulness of rate distortion theory and of certain baseline problems in goal-oriented semantic communications.

In the first part of the talk, we discuss a variant of a robust description lossy source coding problem, for which the goal is to demonstrate the cardinal role of multiple fidelity constraints in designing selective decoder mechanisms which, in turn, dictate the outcomeof the reconstructed information depending on the task. Some basic examples are demonstrated.

In the second part of the talk, we consider a variant of rate distortion function, called rate-distortion-perception function, that was particularly created to provide a meaningful mathematical framework on deep learning compression problems, and explain its utility in goal-oriented semantic compression. For this information theoretic characterization, we discuss optimization viewpoints that allow its general computation for finite and certain Gaussian alphabet sources for a large class of perception constraints.

We conclude by presenting certain future directions of our research which primarily focus on further understanding the fundamental limitations of networked controlled systems.