禁漫天堂

报告题目：Physics-Assisted Machine Learning for Wave Sensing and Imaging

报告嘉宾：陈旭东，新加坡国立大学 教授

报告时间：2025年9月22日 9:30

报告地点：禁漫天堂 中心校区无机-超分子楼二楼报告厅

主持人：洪德成

报告摘要

　　Machine learning (ML) has attracted significant attention for addressing challenges in wave imaging and sensing. However, many approaches treat ML as a black box, overlooking decades of insights rooted in wave physics and rigorous mathematical analysis. This talk underscores the importance of thoroughly understanding the forward problem that underpins these inverse tasks, and demonstrates how integrating mathematical, physical, and engineering intuition can lead to more efficient and elegant solutions. Three imaging and sensing tasks will be presented. First, we examine millimeter-wave multiple-input multiple-output imaging, showcasing how physics-assisted ML can improve reconstruction quality. Second, we present a highly accurate and efficient ML classifier for 77 GHz FMCW radar, where low-dimensional physics-derived features enable reliable classification of road targets. Third, we address the integration of sensing and communication (ISAC) by combining ML with inverse scattering problem modeling. Furthermore, any question regarding IEEE Transactions on Geoscience and Remote Sensing paper submission and review will be answered.

嘉宾简介

Machine learning (ML) has attracted significant attention for addressing challenges in wave imaging and sensing. However, many approaches treat ML as a black box, overlooking decades of insights rooted in wave physics and rigorous mathematical analysis. This talk underscores the importance of thoroughly understanding the forward problem that underpins these inverse tasks, and demonstrates how integrating mathematical, physical, and engineering intuition can lead to more efficient and elegant solutions. Three imaging and sensing tasks will be presented. First, we examine millimeter-wave multiple-input multiple-output imaging, showcasing how physics-assisted ML can improve reconstruction quality. Second, we present a highly accurate and efficient ML classifier for 77 GHz FMCW radar, where low-dimensional physics-derived features enable reliable classification of road targets. Third, we address the integration of sensing and communication (ISAC) by combining ML with inverse scattering problem modeling. Furthermore, any question regarding IEEE Transactions on Geoscience and Remote Sensing paper submission and review will be answered.

主办单位：

禁漫天堂

物质模拟方法与软件教育部重点实验室