題目: Bilevel Additive Models

講座專(zhuān)家：陳洪

時(shí)間：2024,4,18 19:00-21:00

騰訊會(huì)議：583-164-902

摘要: As an important paradigm in statistical machine learning, additive models often exhibit excellent capabilities on function approximation and variable selection. This report will explore the construction and algorithmic implementation of bilevel additive models, with focus on three issues including: (1) How to realize the data-driven structure discovery of variable groups? (2) How to automatically design the appropriate loss function? (3) How to mitigate the impact of noisy features on manifold learning? In theory, the report analyzes the upper bounds of generalization error and the consistency of variable selection. In applications, the effectiveness of bilevel additive models has been validated through data experiments.

報(bào)告人簡(jiǎn)介：陳洪，華中農(nóng)業(yè)大學(xué)教授，博士生導(dǎo)師。研究方向?yàn)闄C(jī)器學(xué)習(xí)，人工智能的數(shù)學(xué)模型與算法。 主持國(guó)家級(jí)項(xiàng)目6項(xiàng)，其中面上項(xiàng)目3項(xiàng)，在人工智能頂會(huì)NeurIPS、ICML、ICLR等發(fā)表論文22篇, 在ACHA、JAT、IEEE TPAMI/TIP/TNNLS/TCYB、Neural Computation、Neural Networks、Pattern Recognition等應(yīng)用數(shù)學(xué)與信息主流期刊發(fā)表論文40余篇。