南京大学计算机软件新技术国家重点实验室

'

摘 要：
Few-shot learning is a learning mechanism that tries to learn and understand new concepts (or categories) from only one or few examples. Humans can learn new concepts with very few instances, and have a strong generalization capability for their variants. Unfortunately, many current machine learning algorithms do not have such a strong generalization ability to identify a new category. Moreover, in some real applications, new samples from new categories are usually difficult to obtain. It is even more difficult to make annotations in many applications. Therefore, learning new categories with very few samples becomes an urgent and important problem. In this talk, we will first give a short review of the advances of few-shot learning, and then introduce our some works on this field.
报告人简介：
李文斌，博士，南京大学计算机科学与技术系助理研究员。2019年于南京大学计算机科学与技术系获博士学位，2013年于中国矿业大学计算机学院获学士学位，并于2017年在美国University of  Rochester联合培养18个月。2020年获江苏省计算机学会优秀博士学位论文奖，2021年获南京大学紫金学者荣誉称号。研究方向为机器学习和计算机视觉，具体包括度量学习，小样本学习，对抗学习和生成对抗网络等。目前在相关研究领域的国际期刊与会议上发表论文20余篇，包括CVPR、AAAI、IJCAI、Neural Networks、MICCAI等。担任IJCV、Neural Networks、Pattern Recognition、ICML、NeurIPS、CVPR、AAAI、Pacific Graphics等多个一流期刊的审稿人和会议的程序委员，并获NeurIPS 2020 “Top 10% high-scoring Reviewers”荣誉称号。

时间：3月25日  12:30-13:40