IoT & AI Joint Seminar | Deep Learning-based Image Compression and Human Action Recognition
Supporting the below United Nations Sustainable Development Goals:支持以下聯合國可持續發展目標：支持以下联合国可持续发展目标：
In this talk, I will summarize my work on deep learning-based image compression (LIC) and human action recognition. Traditional image compression was centered around linear transform, with roots in Digital Signal Processing (DSP). In the last decade, this area has been revolutionized by deep learning, and the linear transform is replaced by powerful neural networks
. My group has proposed some schemes with state-of-the-art performances, and submitted some of the earliest proposals to new image coding standards JPEG AI and IEEE 1857. In particular, we introduced wavelet transform to LIC, which bridges the gap between the DSP and deep learning approaches. It can also be used in other computer vision tasks. In the second topic, I will introduce the privacy-preserving activity sensor that my startup company AltumView has developed for senior care, which was a CES 2021 Innovation Award Honoree and has been in the markets of more than 10 countries.
Jie Liang received the B.E. and M.E. degrees from Xi'an Jiaotong University, M.E. degree from National University of Singapore, and PhD degree from the Johns Hopkins University. Since 2004, he has been with the School of Engineering Science, Simon Fraser University, Canada. His research interests include Signal Processing, Computer Vision, and Machine Learning, with focuses on Image/Video Compression, and Human Action Recognition for Healthcare. He is a Fellow of the Canadian Academy of Engineering, and served as the Associate Director of the School. He was also an Associate Editor for three IEEE Transactions, and a member of three IEEE Technical Committees.