报告题目：Task-Driven Tensor Low-rank Representation in Machine Learning

报告题目：Task-Driven Tensor Low-rank Representation in Machine Learning

报告摘要：Many machine learning tasks can be modeled as a low-rank tensor representation problem. Tensors are the extension of matrices, which represent the multi-way arrays. Different from the matrix, tensor rank is not well defined with a tight convex relaxation. To this end, many formulations of tensor ranks are proposed like CP rank, Tucker rank, Tubal rank, and so on. Although these tensor ranks have their own advantages, they are all designed for general purposes. We argue that for a specific task, we should design a task-tailored tensor low-rank norm to better solve that task. Based on this idea, in this talk, I will present several examples of how to design task-tailored tensor low-rank norms by exploiting the characteristics of specific tasks. Those examples include hyperspectral image representation, subspace clustering, multiple view clustering, etc.

个人介绍：

贾育衡，博士，副教授，江苏省“双创博士”，东南大学“至善青年学者”、“紫金青年学者”。2019年获得香港城市大学(CityU)博士学位，2019-2020年任香港城市大学博士后研究员。导师为Sam Kwong 讲席教授。2020年起在东南大学计算机科学与工程学院任职副教授。现为东南大学PALM实验室成员。曾任斯坦福大学（Stanford University）访问学者（2018年）。研究内容广泛涉及机器学习和数据表示的多个子领域，主要包括半监督学习、高维数据分析与建模、张量表示与建模、图机器学习，深度学习以及在计算机视觉、高光谱表示、社区检测等方向的一些应用。在相关研究领域的国际会议和期刊上发表学术论文40+篇，其中CCF-A/IEEE Trans 论文25+。担任多个国际著名期刊会议的程序委员会委员和审稿人。主持多项国家自然科学基金、江苏省自然科学基金等横纵向项目。