题　目：Recent topics in mixed-integer optimization

嘉　宾：徐立鼎, 博士后, Zuse-Institut Berlin

主持人：葛冬冬 教授 上海交通大学安泰经济与管理学院

时　间：2025年12月15日（周一）14:00-15:30

地　点：徐汇校区安泰浩然308室

内容简介:

In this talk, I will share several research progress in mixed-integer optimization (MIO) that may influence the next generation MIO solvers. The first topic is a comparison between Google’s AlphaEvolve framework and the classical modeling–optimization pipeline. The second topic concerns GPU-accelerations for within MIP solvers. The third topic builds on the relationship between cutting planes, surrogate model, and enumeration. We explored the integration of the Fenchel-cut framework into SCIP solver, which was also named as local cuts in TSP context. Our preliminary results suggest that Fenchel cuts can be stronger than mixed-integer rounding cuts, although their generation is computationally expensive. These findings raise further theoretical questions, particularly regarding how to break the “single-row barrier” that underlies most modern cutting-plane systems.

演讲人简介：

I am a postdoc researcher in IOL group at Zuse-Institut Berlin. I got my Ph.D. at Laboratoire d’informatique de l’École Polytechnique (LIX), where I was part of Optimix.

My work leverages mathematical tools from optimization theory, including lifting techniques, decomposition methods, and cutting-plane methods. These tools enable the design of reliable algorithms for hard optimization problems arising in combinatorics, operations research, and quantum physics.

Beyond theory, I am passionate about translating fundamental advances into practice—developing algorithms that can tackle challenging real-world problems while remaining reproducible and accessible. This motivation naturally connects my research to the development of robust optimization software systems, most notably through my contributions to the open source MILP solver: SCIP.

I primarily work in C++ and Julia. Source codes for my research are usually available in my GitHub repositories.

欢迎广大师生参加!