Yi Tian Xu

Projects

Pattern completion

This project relates to the work I did in my master thesis at McGill University. We can look at certain learning problems as associating the partial input signal pattern to the larger (or complete) signal pattern learned by the agent. In this view, we can not only retrieve the best output to the task in question but also other information related to the input signal.

Visit here for a more complete idea of pattern completion.

So far, this idea has been tested on Human3.6M dataset in this paper and supplimentary results, and on MNIST dataset in this repo.

Respiration sensing using WiFi

WiFi signal reflected off a static human chest has a periodic pattern that is correlated with the respiration rate. With sophisticated signal processing tools and data-driven classification, we can push the sensing accuracy in a noisy environment to a higher level. This project is done at Samsung AI Center, and can be viewed in this paper.

Demo

Sample Artworks

Publications

Conference papers