报告题目：Advances in Semantic Communications based on AI

报告人：Ming Xiao

邀请人：成人动漫与信息交叉团队

时间：2026年7月9日上午10点

地点：成人动漫 犀浦校区七号教学楼7510

报告摘要：

Semantic communication aims to transmit task-relevant meaning rather than exact symbols, thereby improving communication efficiency under limited bandwidth and noisy wireless channels. This talk presents two recent topics that explore robust and efficient semantic communication from different perspectives.

Topic 1: we focuse on multi-modal information bottleneck-based semantic communication. For multi-modal data such as text, audio, and visual signals, different modalities often contain both complementary and redundant information. To address this, we develop a redundancy-aware multi-modal task-oriented communication framework based on a two-stage variational information bottleneck. Uni-modal bottlenecks first extract compact task-relevant representations from each modality, while an adversarial mutual information minimization module suppresses cross-modal redundancy. A second multi-modal bottleneck then produces a channel-robust fused representation for task inference. This design improves both communication efficiency and robustness under channel distortion.

Topic 2, we extend semantic communication to large language model (LLM)-based systems through SkillCom. Instead of using a monolithic LLM pipeline, SkillCom decomposes semantic communication into four explicit skills: semantic abstraction, channel-adaptive transmission, receiver-side repair, and task execution. These skills interact through structured semantic units, allowing channel errors to be localized, repaired, and diagnosed more effectively. Furthermore, SkillCom introduces self-evolving skill optimization to improve skill design and adapt skill selection to different inputs and channel conditions.

Together, these two projects show how semantic communication can move from compact neural representation learning to modular and adaptive LLM-based communication systems, providing more efficient, robust, and interpretable solutions for future intelligent wireless networks.

报告人简介：

Ming Xiao received Ph.D degree from Chalmers University of technology, Sweden in November 2007. From November 2007 to now, he has been in the Department of Information Science and Engineering, school of electrical engineering and computer science, Royal Institute of Technology, Sweden, where he is currently a Full Professor. He was an Editor for IEEE Transactions on Communications (2012-2017), and an Editor for IEEE Transactions on Wireless Communications from 2018 to 2025, and an Area Editor for IEEE Open Journal of the Communication Society from 2019 to 2024. He was Guest Editor for IEEE JSAC (Journal on Selected Areas in Communications) in 2017 and 2024. He has been an Area Editor for IEEE Transactions on Communications since 2026. He is TPC Co-chair for IEEE International Conference on Communications (ICC) 2028. He received IEEE Vehicular Technology Society Best Magazine Paper Award 2023. He is presently a distinguished lecturer of IEEE Communication Society.