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...
演讲人:陈昱安 [北京大学]时间: 16:00-17:00, Oct 21, 2025 (Tue)地点:RM S327, MMW Building内容:The Kitaev toric code remains a benchmark for fault-tolerant quantum computation, yet standard techniques for increasing its logical dimension—lattice surgery, punctures, or concatenation—incur substantial qubit overhead. I will present a unified construction and analysis framework that alleviates ...
演讲人: Benjamin Huard [Ecole Normale Supérieure de Lyon]时间:10:00-11:30, Oct 17, 2025 (Fri)地点:RM S527, MMW Building内容:Is it possible to count the number of photons in a microwave cavity by exploiting the dispersive interaction between a superconducting qubit and the cavity? In standard protocols, this operation requires to perform a series of gates on the qubit that are conditioned on...
演讲人: 冯昱善 [UT Austin]时间: 16:30-18:00, Sep 25, 2025 (Thu)地点:RM 1-222, FIT Building内容:Much of the current progress of AI is driven by massive datasets, unprecedented compute, and industrial-scale engineering. Yet for many of us in academia, the path forward need not be defined by scale alone. This talk explores a complementary direction: uncovering new insights by reinterpreting th...
演讲人:Tonghan Wang [Harvard University] 时间: 15:00-16:00, Sep 25, 2025 (Thu)地点:RM 1-222, FIT Building内容:The ecosystem of Al agents and humans will shape the future. For the success of thisecosystem, we need to build agents and agent systems that can scale to real-worlccomplexity while providing alignment guarantees. This goal necessitates new progressin multi-agent Al, This talk prese...
演讲人: Haozhi Qi [UChicago/UC Berkeley]时间: 16:00-17:30, Sep 23, 2025 (Tue)地点:Seminar Room 2, 19th Floor, Tower C, TusPark (#腾讯会议:398-077-2882 密码:095015)内容:Human hands are essential for sensing and interacting with the physical world, enabling us to manipulate objects with ease. Replicating this level of dexterity in robots remains a longstanding challenge, and a key milestone...