Distortion Resilience for Goal-Oriented Semantic Communication

TL;DR

This work tackles how to achieve efficient, goal-oriented semantic communication under wireless distortions by reframing semantic metrics as data distortions analyzed through rate-distortion theory. It introduces DRGO, a distortion-driven framework that precomputes AI task performance from distortion to guide joint compression and resource allocation while minimizing energy. To solve the resulting non-convex optimization, the authors develop DDPG-EI, a deep reinforcement learning agent that enforces explicit constraints and handles implicit constraints via a penalty mechanism, with a state/action/reward design tailored to semantic distortion and AI performance. Theoretical bounds connect data distortion to AI training and inference performance, and experiments on CIFAR-10 with multi-user settings demonstrate significant improvements in transmission efficiency and task accuracy under distortion constraints, highlighting the practical value of online, data-driven optimization for SemCom in IoE environments.

Abstract

Recent research efforts on Semantic Communication (SemCom) have mostly considered accuracy as a main problem for optimizing goal-oriented communication systems. However, these approaches introduce a paradox: the accuracy of Artificial Intelligence (AI) tasks should naturally emerge through training rather than being dictated by network constraints. Acknowledging this dilemma, this work introduces an innovative approach that leverages the rate distortion theory to analyze distortions induced by communication and compression, thereby analyzing the learning process. Specifically, we examine the distribution shift between the original data and the distorted data, thus assessing its impact on the AI model's performance. Founding upon this analysis, we can preemptively estimate the empirical accuracy of AI tasks, making the goal-oriented SemCom problem feasible. To achieve this objective, we present the theoretical foundation of our approach, accompanied by simulations and experiments that demonstrate its effectiveness. The experimental results indicate that our proposed method enables accurate AI task performance while adhering to network constraints, establishing it as a valuable contribution to the field of signal processing. Furthermore, this work advances research in goal-oriented SemCom and highlights the significance of data-driven approaches in optimizing the performance of intelligent systems.

Figures (12)

• Figure 1: Illustration of an image being affected by different distortions via the communication channel. a) the original image, b) image affected by Gaussian noise with variance of 25, c) image affected by spackle noise with variance of 100, d) image affected by Gaussian noise with variance of 250.
• Figure 2: System model of the considered SemCom.
• Figure 3: The key features of DDPG-EI are the action settings: the actions belong to explicit constraints from actor network are scaled using Sigmoid, Softmax or ReLU functions according to specific explicit constraints for computational complexity reduction.
• Figure 4: DRL performance on training optimization problem.
• Figure 5: Illustration of the inference generalization gap (a) with different decision boundaries, and (b) with different model accuracy.

Theorems & Definitions (14)

• Example 1: Distorted Data Impact on AI Performance
• Definition 1: AI Inference
• Definition 2: AI Training
• Lemma 1
• Theorem 1
• proof
• Lemma 2
• Theorem 2: Model training degradation evaluation
• Lemma 3: Total variation of distorted data
• Theorem 3: Model inference degradation evaluation