-
09-14
2026
Designing AI Infrastructure as an Integrated Whole
时 间：10:00-11:00, Sep 14, 2026 (Mon)地 点：https://meeting.tencent.com/dm/kMlAAN2VkyZ8 #腾讯会议：554-221-000内容：Modern AI infrastructure combines specialized training runtimes, GPU kernels, and communication libraries. Optimizing these components individually, however, does not necessarily meet end-to-end requirements for performance and reliability. My research examines these requirem...
-
09-09
2026
State retrieval beyond Bayes’ retrodiction and reverse processes
时 间：15:00-17:00, Sep 9, 2026 (Wed)地 点：MMW-727会议室 (腾讯会议： 422-776-136)内容：In the context of irreversible dynamics, the meaning of the reverse of a physical evolution can be quite ambiguous. It is a standard choice to define the reverse process using Bayes' theorem, but, in general, this is not optimal with respect to the relative entropy of recovery. In this work we explore wh...
-
09-08
2026
通过自我演进的操作系统层虚拟化大模型智能体
时 间：10:00-12:00, Sep 8, 2026 (Tue)地 点：RM 1-202, FIT Building内容：AI 应用已经从单一、单体的基础模型（FM）转向复合型智能体系统。然而，今天的技术栈仍然是碎片化的：即便协议（如 MCP、A2A）缓解了工具/智能体连接问题，每个框架仍内嵌了一个隐式运行时，用于处理状态、记忆、预算和护栏，这使得行为不可移植，治理也变得脆弱。这与操作系统出现之前的计算环境如出一辙，那时每个程序都要重新实现基本服务。本...
-
08-25
2026
Detecting quantum resources in mixed states
时 间：11:00-12:00, Aug 13, 2026 (Thu)地 点：RM S527, MMW Building内容：Detecting quantum resources, especially in mixed quantum states, is a highly challenging task. Starting from the widely-used concept of resource witnesses, we propose a framework that systematically generate analytical and experimentally accessible criteria based on a single witness and a group of free unitaries.Our co...
-
08-21
2026
Global Dualities in Quantum Integrable Systems and Markov Processes
时 间：11:00-12:00, Aug 21, 2026 (Fri)地 点：腾讯会议：188-585-149（会议链接：https://meeting.tencent.com/dm/ueSfCeE8ENCB）内容：Duality transformations reveal unexpected equivalences between seemingly distinct models. We introduce a concept of global duality, in which a duality operator intertwines two one-dimensional local Hamiltonians via generalised exchange relations, in contrast to lo...
-
08-07
2026
Video Generation Models are General-Purpose Vision Learners
时 间：10:00-11:00, Aug 7, 2026 (Fri)地 点：https://meeting.tencent.com/dm/z9BkdA8pj32P (https://meeting.tencent.com/dm/z9BkdA8pj32P)内容：Driven by next-token prediction, NLP shifted from task-specific models into powerful generalist foundation models. What, then, is the equivalent catalyst needed to achieve a general-purpose model in computer vision?In this talk, we contend that large-sca...