Self-Supervised Spectral Imaging: From Remote Sensing to Materials Science
数学专题报告
报告题目(Title)：Self-Supervised Spectral Imaging: From Remote Sensing to Materials Science
报告人(Speaker)：王超（南方科技大学）
地点(Place)：腾讯会议：200-607-813 会议密码：251112
时间(Time)：2025年11月12日（周三）16:00-17:00
邀请人(Inviter)：王发强
报告摘要
Spectral imaging, including hyperspectral imaging, presents significant challenges in machine learning and computer vision due to high data dimensionality and a scarcity of labeled ground truth. This talk addresses these challenges by introducing a unified, self-supervised learning framework applicable across diverse domains, from remote sensing to materials science. The proposed methodology synergistically combines low-rank matrix factorization, continuous neural representations, and variational inference techniques derived from a model-based perspective. This integrated approach effectively learns robust representations without relying on extensive annotated datasets, demonstrating its versatility and power in practical applications.
主讲人简介
王超，南方科技大学统计与数据科学系副研究员，博导，其研究方向主要为图像处理、科学计算与交叉学科的数据科学。在本领域期刊SIAM系列、IEEE汇刊等杂志及学术会议发表学术论文三十余篇。在2022年CVPR研讨会获得最佳论文，在2021年获深圳市鹏城孔雀计划特聘岗位，在2017年获得中国工业与应用数学学会年会最佳论文。主持国自然青年基金、广东省面上基金以及深圳市稳定支持面上项目，以课题负责人或核心成员参与国家重点研发项目、香港研资局科研基金项目以及深圳重点项目。
