As a new media type, micro-videos allow users to record their daily life within a few seconds and share over social media platforms. Recently, the micro-video is explosively growing with the rapid development of network technology and intelligent devices. Therefore, how to use the automatic and intelligent methods for the micro-video analyzing and understanding has become an urgent problem in the academia and industry.

Led by professor Liqiang Nie, iLearn explores the problems and challenges in the micro-video analysis and understanding. We first build a large-scale micro-video dataset for some real-world applications. And then, we extend and improve the existing techniques, such as deep transfer learning, cooperative learning and dictionary learning. These works cover some key problems in the micro-video, including multimodal and heterogeneous signals, insufficient information, and low quality. Besides, we explore the complex relationships among the micro-video, user, and some social information. With the help of graph convolutional networks, we propose some powerful frameworks for the applications.

Following the works presented by Prof. Nie during his post-doctoral period, iLearn keeps further exploring the micro-video analysis and understanding. Until now, more than 10 works have been published in top forums, like ACM MM, SIGIR and IEEE Transactions on Image Processing. In particular, one academic monograph titled “Multimodal Learning Toward micro-video Understanding” has been published by Morgan & Claypool Publishers. And, more than 10 national patents are authorized or accepted by the National Intellectual Property Administration. Meanwhile, the works are supported by the National Natural Science Foundation of China and the Tencent AI Lab Rhino-Bird Joint Research Program. Moreover, we keep deeply cooperation with some well-known universities and research institutions around the world, such as the National University of Singapore, Kwai, and Tencent AI Lab.