Neural Topic Model with Attention for Supervised Learning (AISTATS 2020, long paper)
Xinyi Wang, Yi Yang.
Jun 2019 - Oct 2019
I continued to work with professor Yi Yang as a Junior Research Assistant this summer. In this new project, we bring the supervised deep learning model and topic model together by designing a novel attention mechanism. After conducting comprehensive experiments on three public datasets with different types of labels, we conclude that our proposed topic attention model significantly outperforms the baselines on both the supervised tasks and perplexity.
Applied Mathematics in Fluid Statics
Direct proof of the formation of droplet surface shape and the principle of minimizing free energy (Under review at Acta Physica Sinica)
Kang Jin, Xinyi Wang, Kaihang Gui.
Jun 2017 - Aug 2017, Oct 2019
Under the guidance of Prof. Xiaoping Wang, I elaborated on some existing findings of the static liquid behaviors on solid surfaces under the University Research Opportunity Program (UROP) of HKUST. I unintentionally derived a direct proof of that, ideally, the surface of a droplet is a sphere under the principle of minimizing free energy, using the calculus of variation and Lagrange multiplier. (i.e. a simplified version of the isoperimetric inequality in R3.) An interesting follow-up is that, as I posted my proof online, two years later, Dr. Kang Jin from Northwest University (China) contacted me about using it in his publication.
