Exploration and Application on Compact Embedding Space in Domain Generalization

The School of ITEE is hosting the following thesis confirmation seminar:

Exploration and Application on Compact Embedding Space in Domain Generalization

Speaker: Kaiyu Guo
Host: Prof Brian Lovell

Abstract: With the development of deep learning, many computer vision tasks have achieved astonishing progress, such as image classification and image detection. Actually, there are many deep learning algorithms having been applied in our daily life, like face recognition, which means the performance of such methods has reached a reliable level. However, relying on the training data, the deep learning algorithms are hard to deal with the gap between different domains which we can call domain shift. There are many tasks have been proposed to solve the domain shift issues like domain adaptation and domain generalization. This research aims at exploring the feature distribution in the embedding space of multi-domain situations. We will focus on the compact embedding space in domain-generalization task which is a research topic on the domain shift between source domains and unseen domains. Unlike previous algorithms focusing on domain alignment in representation learning, we propose our algorithms based on the motivation of weakening the domain information during the training process. In addition, we will also focus on the usage of the cross-domain compact embedding space, such as the new research topic of cross-domain incremental learning. This report also presents the outcome we have achieved till now and the timeline designed for the thesis. We already have a conference submission for our first project of domain-aware triplet loss.

Speaker bio: Kaiyu Guo is a  Ph.D. candidate from the School of ITEE at the University of Queensland under the supervision of Prof Brian Lovell.  Kaiyu received his bachelor degree in mathematics from the Dalian University of Technology, China, and his master degree in computing mathematics from the University of Science and Technology of China. His research interests include computer vision algorithms, domain generalization, and contrastive learning.