A cross-domain lifelong learning model for visual understanding

Abstract

In the study of media machine perception on image and video, people expect the machine to have the ability of lifelong learning like human. This paper, starting from anthropomorphic media perception, researches the multi-media perception which is based on lifelong machine learning. An ideal lifelong machine learning system for visual understanding is expected to learn relevant tasks from one or more domains continuously. However, most existing lifelong learning algorithms do not focus on the domain shift among tasks. In this work, we propose a novel cross-domain lifelong learning model (CD-LLM) to address the domain shift problem on visual understanding. The main idea is to generate a low-dimensional common subspace which captures domain invariable properties by embedding Grassmann manifold into tasks subspaces. With the low-dimensional common subspace, tasks can be projected and then model learning is performed. Extensive experiments are conducted on competitive cross-domain dataset. The results show the effectiveness and efficiency of the proposed algorithm on competitive cross-domain visual tasks.