Joint Data and Semantics Lossy Compression: Nonasymptotic and Second-Order Achievability Bounds

Author:

Huiyuan Yang, Yuxuan Shi, Shuo Shao, Xiaojun Yuan

Keyword:

Computer Science, Information Theory, Information Theory (cs.IT)

journal:

date:

2024-01-26 00:00:00

Abstract

This paper studies a joint data and semantics lossy compression problem in the finite blocklength regime, where the data and semantic sources are correlated, and only the data source can be observed by the encoder. We first introduce an information-theoretic nonasymptotic analysis framework to investigate the nonasymptotic fundamental limits of our studied problem. Within this framework, general nonasymptotic achievability bounds valid for general sources and distortion measures are derived. Moreover, we provide a second-order achievability bound in the standard block coding setting by applying the two-dimensional Berry-Esseen theorem to our nonasymptotic bounds. Compared with first-order asymptotic bounds, our results have the potential to provide unique insights for the design of practical semantic communication systems.

PDF: Joint Data and Semantics Lossy Compression: Nonasymptotic and Second-Order Achievability Bounds.pdf