题目：

Forget BIT, It is All about TOKEN: Towards the Semantic Information Theory for LLM

摘要：

近年来，以大模型为代表的AI技术发展迅猛，并开始在许多领域发挥重要作用。然而，对于大模型背后原理的研究还比较少。实际上，人们对大模型理论理解的欠缺也限制该技术在许多关键场景中的应用。本次报告我们从大模型时代的核心概念-TOKEN出发，借鉴香农信息论的思想和方法论，尝试建立一个理解大模型原理的语义信息论框架。报告的主要内容包括：语义的向量表示及压缩，大模型的概率建模，参数化记忆和大模型的语义信息流等。我们期望这些尝试能启发研究人员关注面向AI时代的新信息理论，期待在面向大模型的语义信息论方面做出原创性成果。

报告人简介：

题目：

On the Fundamental Limits of Generative Communication

摘要：

Motivated by the emerging paradigm of generative communication, this talk explores the problem of channel-aware optimal transport, where a block of i.i.d. random variables is transmitted through a memoryless channel to generate another block of i.i.d. random variables with a prescribed marginal distribution such that the end-to-end distortion is minimized. With unlimited common randomness available to the encoder and decoder, the source-channel separation architecture is shown to be asymptotically optimal as the blocklength approaches infinity. On the other hand, in the absence of common randomness, the source-channel separation architecture is generally suboptimal. For this scenario, a hybrid coding scheme is proposed, which partially retains the generative capabilities of the given channel while enabling reliable transmission of digital information. It is demonstrated that the proposed hybrid coding scheme can outperform both separation-based and uncoded schemes.

报告人简介：

题目：

A Unified View on the Decoding of Reed-Solomon Codes

摘要：

Reed-Solomon (RS) codes have been widely applied in data storage and transmissions, thanking to their maximum distance separable (MDS) property that ensures their high and robust error-correction capability. Its fine algebraic structure also enables rich decoding methodologies. This talk provides a unified view on the algebraic decoding of RS codes, showing their solutions can be founded through computing the Gröbner bases. This unified understanding may inspire further enterprising endeavor for developing more advanced RS decoding.

报告人简介：

题目：

Convergence of Federated Learning with Communication Distortions

摘要：

Federated learning (FL) becomes increasingly attractive in the areas of wireless communications and machine learning due to its powerful learning ability and potential applications. In contrast to other machine learning techniques that seldom require communication resources, FL exploits communications between the central server and the distributed local clients to train and optimize a model. Therefore, how to efficiently assign limited communication resources to train a FL model is critical to performance optimization. In this talk, we investigate the convergence performance of FL under various communication distortions. After introducing the fundamental tradeoff between efficiency and convergence, we discuss the convergence of FL via inexact ADMM and decentralized FL.

报告人简介：

题目：

面向通信信息系统的概率神经网络优化

摘要：

在新一代通信信息系统中，复杂信道环境、动态资源调度以及多样化业务需求对系统建模与优化提出了前所未有的挑战。深度学习为通信信息系统提供了新的解决思路，但在直接应用中仍存在诸多不足，例如模型可解释性差、泛化能力有限以及不确定性建模不完善等。针对这些问题，团队长期开展了基于概率建模的深度神经网络优化研究，提出了从非高斯分布建模、变分推断方法到贝叶斯学习框架在内的多种概率推断与建模策略，并应用于正则化、注意力机制解释、不确定性估计和生成式优化等核心环节。相关研究不仅有效缓解了模型的过拟合与训练偏差，还提升了其在小样本学习、跨域迁移与多模态理解中的泛化能力与可解释性。更重要的是，这些方法在信号检测、可靠传输、资源分配与系统性能优化等通信关键环节展现出良好应用前景，为智能通信信息系统的构建提供了新的理论与方法支撑。

报告人简介：

题目：

我们与华为的合作及其他

摘要：

我将从我们与华为的合作谈起，介绍近年来我们在数学与交叉领域研究方向的一些历程和研究成果。同时也将简单谈谈我对人工智能的一些看法。

报告人简介：

题目：

Recent Progress in Polar Coding

摘要：

This talk presents the latest advancements in polar coding. We propose novel decoders featuring enhancements targeting both higher performance and lower complexity. For improved reliability, we demonstrate algorithms with minimal hardware overhead; for the reduced complexity, we studied the inherent advantage of polar codes to support the ultra-high-throughput and low-power applications. On the encoder side, we explore how polar codes benefit from a novel stitching structure to obtain significant performance gains. Furthermore, for incremental redundancy (IR) HARQ design, we present the breakthrough of polar codes to generate multiple redundancy versions for flexible retransmissions. We also surveyed the latest advancements from top research teams in China and abroad. Their findings may change the prevailing opinion about polar codes, for example, by showing that polar codes can outperform LDPC codes even at large block lengths and with low complexity. Overall, these advancements demonstrate that polar codes have the advantage to address the diverse requirements of wireless applications.

报告人简介：

题目：

A Brief Introduction to Secure Multiparty Computation

摘要：

Secure multi-party computation (also known as secure computation, multi-party computation (MPC) or privacy-preserving computation) is a subfield of cryptography with the goal of creating methods for parties to jointly compute a function over their inputs while keeping those inputs private. In this talk, I will briefly introduce this topic and some crypto primitives. 

报告人简介：