时 间：2026-05-23 09:30-10:30
地 点：普陀校区理科大楼A1514室
报告人：徐茂超 伊利诺伊州立大学教授
主持人：陈律 华东师范大学副教授
摘 要：
<img class="akeylayout_img"r breaches continue to pose significant threats to enterprises and society, necessitating robust models for accurate risk assessment and effective insurance pricing. Traditional approaches in industry typically rely on offline model updates, which are timeconsuming, costly, and cannot be performed regularly at short intervals; as a result, models are often refreshed only infrequently, delaying the incorporation of new information. In this work, we introduce a novel online learning framework that integrates copula-based dependence modeling with natural gradient boosting techniques to capture both temporal and cross-group dependencies in cyber breach data. Our method continuously updates model parameters as new data becomes available, leveraging online linearization to efficiently adjust to evolving risk profiles. This adaptive framework employs D-vine and bivariate copula structures to flexibly model non-linear dependencies among risk factors while incrementally refining marginal distributions through online updates. Empirical evaluation of the real-world cyber incident dataset demonstrates that our approach not only captures the intricate statistical characteristics of multivariate cyber risks but also outperforms conventional methods in predictive performance.
报告人简介：
Dr. Maochao Xu is a Full Professor of Mathematics at Illinois State University. His research focuses on cyber risk, statistical modeling, and deep learning. He has published numerous papers in prestigious journals, and his work has been recognized with distinctions such as the Best Paper Award in insurance, and Featured Articles in cybersecurity. In addition to his academic contributions, he provides advisory and consulting services to industry leaders in cyber insurance.