题  目：深度谱核网（Deep Spectral Kernel Networks）

报告人:  薛晖（东南大学）

报告摘要

As more competitive approaches than neural networks and kernel methods, deep spectral kernel networks (DSKN) are revolutionizing representation learning by offering unique benefits: more powerful representation ability and excellent efficiency in computational elements. These networks have been successfully used in many complex tasks, including 3D view synthesis, time series encoding, etc. In this talk, we will systematically present our latest research works about DSKN: 1) the framework of deep spectral kernel networks; 2) a new state-of-the-art parameter-modeling approach solving the optimization dilemma; 3) a tremendous soft transfer scheme extending DSKN to more than 100 layers; 4) a set of rectified continuous Bernoulli units to automatically gate multi-frequency patterns; 5) a solid application method for time series domain adaptation based on DSKN. Finally, we will discuss profound thinking on further development of deep spectral kernel networks.

报告人简介: 薛晖，女，东南大学计算机科学与工程学院、软件学院、人工智能学院教授，博士生导师。主要研究领域包括机器学习与模式识别。在《IEEE Transactions on Pattern Analysis and Machine Learning》、《IEEE Transactions on Neural Networks and Learning Systems》、AAAI、IJCAI、CVPR、ACM MM等重要国际期刊和领域顶级会议上发表论文60余篇，主持多项国家自然科学基金项目和江苏省自然科学基金项目。现任中国计算机学会人工智能与模式识别专业委员会委员、中国人工智能学会机器学习专业委员会委员、江苏省人工智能学会机器学习专业委员会副主任、江苏省人工智能学会智能系统与应用专业委员会副主任等。

腾讯会议ID： 203-604-848

时     间 :  2022年11月14日（星期一） 下午 13：50 开始

联  系  人： 胡恩良 18788400609

-- 欢迎广大师生参加 --