学术报告1：Evolutionary Multi-objective Feature Selection for Machine Learning
报告人：焦儒旺
工作单位：苏州大学
报告时间：2024年9月20日（星期五）08:00
报告地点：淮南校区明德楼208
报告人简介：焦儒旺，苏州大学青年特聘教授，德国“洪堡学者”，日本学术振兴会海外特别研究员（JSPS学者）。研究方向为演化计算、机器学习和数据挖掘。目前主持国家自然科学基金1项，其它项目3项。已发表学术论文27篇（一作/通讯23篇），其中包括IEEE Transactions on Evolutionary Computation、Evolutionary Computation、IEEE Transactions on Cybernetics和Information Sciences计算智能领域顶刊 12篇。曾获IEEE WCCI国际学术竞赛亚军1项。现担任IEEE智能计算协会Taskforce on Evolutionary Computation for Feature Selection and Construction的主席，2024年CCF中国网络大会组织委员会共同主席，以及两个国际期刊的编委。
报告摘要：Maximizing classification accuracy and minimizing the number of selected features are the two primary objectives in feature selection, making it a multi-objective task. Multi-objective feature selection not only reduces dimensionality and improves accuracy, but also provides insights from complex data. Despite significant advancements, there are still some challenges that need to be addressed, such as complex feature interactions, exponentially growing search space, solution duplication in the objective space, and objective selection bias. In this talk, we will introduce recent works addressing these challenges and highlight emerging directions for the future of multi-objective feature selection.
学术报告2：Boosting SAT-based local improvement methods
报告人：夏海
工作单位：维也纳工业大学
报告时间：2024年9月20日（星期五）08:45
报告地点：淮南校区明德楼208
报告人简介：夏海，维也纳工业大学（TU Wien）算法与复杂度（Algorithms and Complexity）实验室博士生，维也纳逻辑与算法中心研究员，欧盟“玛丽居里”（Marie Skłodowska-Curie Ph.D. fellowship）项目资助学者。研究方向为：布尔可满足性问题求解（SAT），启发式搜索（heuristic search），算法选择（algorithm configuration），演化学习（evolutionary learning）。已发表学术论文六篇，其中包括AAAI，ICTAI等人工智能领域旗舰会议。曾获IEEE WCCI国际学术竞赛亚军1项。
报告摘要：Over the last two decades, SAT-solver technology has made tremendous progress; SAT instances with over a million variables can be solved routinely. However, for many combinatorial optimization problems, the encoding to SAT entails a significant blow-up in size (cubic or worse), significantly limiting the feasible instance size of the combinatorial problem. To make SAT still applicable to large combinatorial problem instances, researchers have developed new algorithmic frameworks where SAT solvers are called multiple times, where each individual SAT call deals only with a small part of the combinatorial instance. SAT-based Local Improvement (SLIM) is such an algorithmic framework where multiple local SAT calls improve a global heuristic solution by encoding and solving local subinstances iteratively. Therefore, we want to broaden the application area of the SLIM framework, where heuristic methods and exact methods can mutually benifit from each other, and we want to configure the SLIM framework in a systematic way, boosting the efficiency of solving different computationally hard combinatorial problem.