报告题目：Structured Feature Selection of Raman Spectroscopy based on Sparse Group LASSO  

报告时间：2024年1月15日（周一）14：00

报告地点：C1-435

主讲人：张寅升 副 教授

报告摘要：

Group structures are common, e.g., gene co-expression in biology, multiple bio-marker existence in the same biochemical pathway, term co-occurrence in natural language patterns, etc. Due to its physio-chemical nature, Raman spectroscopy also possesses such structures, i.e., multiple peaks caused by the same chemical bond or functional group. These inherent structures indicate that the observed features should not be treated as mutually independent or uncorrelated. Therefore, we proposed a sparse group-LASSO feature selection (FS) algorithm augmented by a Raman peak assignment knowledgebase. The algorithm has an L1 penalty that ensures feature sparsity and a structured L2 penalty that favors group selection. In a cheese source identification case study, the proposed method has successfully selected group-structured Raman peak features, and the results are highly interpretable for the target task. Finally, we implemented the algorithm as open-sourced software to benefit the community. Peer re-searchers can also generalize this structured FS method to other spectroscopic profiling modalities.

主讲人简介：张寅升，副教授，硕导。1986年出生，本科至博士就读于浙江大学，CSC访问学者。近年来从事质量安全、检测技术和数据科学相结合的交叉研究方向，发表论文40余篇。近五年以第一作者发表SCI期刊论文10余篇，其中，JCR一区4篇（含中科院一区top论文2篇）。被引200余次。课题研究10余项，其中，国家自然科学基金主持2项、国家重点研发计划子课题主持1项。成果转化方面，获批专利及软著20余项、国家标准1项，发布产业指数1项。相关成果于2021年入选国家自然科学基金委科学传播与成果转化中心推荐名单，并被中国日报、中国经济新闻、文汇报等媒体广泛报道。