报告人：陈小丽（中国地质大学）

报告时间：2026年10月9日（星期五）14:30-16:30

报告地点：科技楼南楼706室

报告摘要：Stochastic delay differential equations (SDDEs) are fundamental mathematical frameworks for modeling complex dynamical systems with intrinsic randomness and time-delayed feedback. However, data-driven identification of SDDEs remains challenging due to the simultaneous estimation of unknown time delays and the drift/diffusion terms, and the robustness of the method under multi-parameter scenarios. To address these limitations, we propose a data-driven identification framework that combines mutual information, Bayesian optimization, and deep learning. It leverages mutual information as a statistically robust objective to quantify dependencies between historical and future states while mitigating noise interference. Bayesian optimization is employed as a gradient-free global search strategy to efficiently locate the optimal delay without requiring a priori information. A dual-branch neural network architecture is further designed to decouple the learning of the drift and diffusion terms, enhancing the approximation accuracy of these components. Furthermore, a multi-parameter mechanism is incorporated, enabling the framework to generalize across a class of systems within a single training process.

报告人简介：陈小丽，中国地质大学（武汉）特任教授。2020年博士毕业于华中科技大学。2018年9月至2020年8月在美国布朗大学进行联合培养。2021年3月至2024年3月在新加坡国立大学从事博士后研究。2024年入选湖北省高层次人才计划，主要从事随机动力系统, 机器学习与动力系统相关研究。已在SIAM Journal on Scientific Computing, Nature Computational Science, CMAME, Physica D等期刊发表多篇学术论文。

邀请人：王季红