政治大學應用數學博士,目前服務於臺灣大學資料科學學位學程,講授Python程式設計課程。研究興趣為幾何重建、深度學習、HodgeRank 的應用。
副職是不學無術者,一開始攻讀投資組合最佳化問題,接著渡過了記憶體設計、奈米尺度材料力學及智慧理財新創公司等業界生活,在"被"投入到教學神經網路的講師狀態後,開始研究微分幾何與神經網路之間的跨領域應用。
Taipei, Taiwan
Dissertation: Manifold Embedding with Deep ResNet and A Study of the Continuity of HodgeRank
I have two main research topics:
a) HodgeRank and its real-world applications;
b) Manifold reconstruction using Neural Network.
For the first topic, I mainly deal with HodgeRank method and its generalization. A sufficient condition of continuity of HodgeRank was studied.
For the second topic, I reformulate a manifold interpolation algorithm as a learning process of deep residual network (ResNet) using the characterization property of affine orthogonal projections.
Studied on manifold interpolation/reconstruction based on composition of local affine projections
Thesis: Stochastic Portfolio Optimization Models for the Stable Growth Benchmark Tracking
I focused on the combination between portfolio optimization problem and scenario trees. Moment matching method was introduced to provide a user-defined scenario tree which fits the first few moments of past data.
Major in Judo (2nd Dan), double major in Karate (brown belt), minor in Slavic languages and applied mathematics. Besides, I am one of the co-founders of speciality coffee club in NCCU.
HodgeRank is a pairwise comparison method for recommendation system.
The theory of HodgeRank is based on the theory of cohomology group and Hodge/Helmholtz Decomposition.
The figure is referred from the paper by Jiang, X., Lim, L. H., Yao, Y., & Ye, Y.
I am recently focus on how to use neural network in medical image analysis. I am working with master students in NTU with Prof. Weichung Wang. We are now trying to duplicate some existence models and then improve it in any sense.
This figure is the MRI image of my left shoulder.