Semantic Communications With Artificial Intelligence Tasks: Reducing Bandwidth Requirements and Improving Artificial Intelligence Task Performance
Yang Yang, Caili Guo, Fangfang Liu, Lunan Sun, Chuanhong Liu, Qizheng Sun
Abstract
A radical paradigm shift of wireless networks from “connected things” to “connected intelligence” is occurring, which coincides with Shannon and Weaver’s vision that communications will transform from the technical level to the semantic level. This article proposes a semantic communication (SC) method with artificial intelligence tasks (AITs). First, the architecture of SC-AIT is elaborated. Then, based on the proposed architecture, we implement SC-AIT for an image classifications task. A prototype of SC-AIT is also established for surface defect detection. Experimental results show that SC-AIT has much lower bandwidth requirements and can achieve up to 40% classification accuracy gains compared with communications at the technical level. Future trends and key challenges for SCs are also identified.