演讲人:刘勇(中国人民大学)时间:11:45-13:00, Nov 27, 2025 (Thu)地点:RM 1-202, FIT Building内容:Transformer 是当前大模型的核心架构,在自然语言处理、计算机视觉等领域展现出卓越性能,但其内在机理的理解仍较为有限。本报告从统计物理能量视角构建新框架,发现自注意力机制本质是最小化 Helmholtz 自由能的过程,由此为其提供了物理层面的原理解释。在此基础上,进一步从能量视角设计了新型 Transformer 架构,且实...
演讲人:Mikkel Thorup [University of Copenhage]时间:14:00-15:00, Nov 24, 2025 (Mon)地点:RM 1-222, FIT Building内容:A clustering is a partitioning into sets called clusters. The input to (unweighted) correlation clustering is an unweighted graph. We want to cluster the vertices so as to minimimize the number of edges between clusters plus the number of non-edges inside clusters.Recently there...
演讲人:Jung Ho Ahn [首尔国立大学] 时间:10:00-11:00, Nov 7, 2025 (Fri)地点:RM 1-222, FIT Building内容:GPUs continue to redefine performance boundaries across diverse workloads. This talk presents two compelling examples from our latest research, underscoring the critical role of programmable accelerators. First, we demonstrate how our group surpassed DARPA's ambitious 2021 DPRIVE ASIC target...
演讲人:Lingjie Duan [SUTD]时间:14:00-15:00, Nov 6, 2025 (Thu)地点:FIT1-312内容:Existing AI methods (model-based optimization, deep reinforcement Learning) struggle with generalization and rapid adaptation in dynamic wireless networks. This work pioneers "LLM for Wireless," harnessing the general intelligence of Large Language Models (LLMs) to proactively optimize network infrastructure, mov...
演讲人:刘鑫 [华中科技大学]时间:16:00-17:00, Oct 28, 2025 (Tue)地点:RM S327, MMW Building内容:Majorana zero modes (MZMs) exhibit non-Abelian braiding statistics, rendering them a cornerstone for topological quantum computing. A robust MZM platform demands a wide topological superconducting parameter space that facilitates experimental detection and control of MZMs and provides a promising de...
演讲人: 吴磊 【北京大学】时间:14:00-15:00, Oct 24, 2025 (Tue)地点:吕大龙楼701会议室内容:大模型尺度律揭示:性能随训练数据规模和计算量的增长呈现出可预测的幂律提升。这一规律深刻推动了现代人工智能的发展,但长期停留在经验观察层面,缺乏理论理解。为探究其成因,我们引入幂律核回归(power-law kernel regression)这一简化模型,作为理论原型来抽象尺度现象的关键机制。基于该模型的动力学推导,我们提出泛函尺...