Learning Multi-Rate Task-Oriented Communications Over Symmetric Discrete Memoryless Channels

TL;DR

This paper proposes a multi-rate task-oriented communication framework that adapts to changing data rates over symmetric discrete memoryless channels, enabling efficient edge inference across variable conditions.

Contribution

It introduces a novel multi-rate framework with a nested codebook and progressive learning, allowing dynamic adaptation to channel variations for task-oriented communication.

Findings

Supports edge inference across multiple coding levels

Effectively adapts to variable communication environments

Outperforms traditional fixed-rate methods

Abstract

This letter introduces a multi-rate task-oriented communication (MR-ToC) framework. This framework dynamically adapts to variations in affordable data rate within the communication pipeline. It conceptualizes communication pipelines as symmetric, discrete, memoryless channels. We employ a progressive learning strategy to train the system, comprising a nested codebook for encoding and task inference. This configuration allows for the adjustment of multiple rate levels in response to evolving channel conditions. The results from our experiments show that this system not only supports edge inference across various coding levels but also excels in adapting to variable communication environments.