About Publications Projects CV



Rui Zhang (张瑞) (I am going to update this webpage slowly, 2020/04.)

I'm a Computer Science grad from Australian National University, interested in Machine Learning & Optimization. I'm currently a member of Computational Media Lab, supervised by Marian-Andrei Rizoiu.

I was an undergrad at Shanghai Jiao Tong University, with a major in Mechanical Engineering . I mainly worked on Robot Motion Planning and Control, supervised by Chenkun Qi. I was a visiting student at Purdue University from Aug to Dec 2015.

Here's my CV. I am planning to pursue a Ph.D.

Recent Publications (None...)

Recent Projects

Learning Non-parameteric Triggering Kernels of Hawkes Point Processes [arXiv][Code](TODO)

Lots of recent work is on strengthing Point Processes via neural networks or Gaussian Process because non-parametric models equip Point Processes with higher fitting capacity. However, they mainly focus on improving the intensity function of Point Processes, as a result of which accurately uncovering branching structures of Point Processes is still hard work. Concerning about this, we use non-parametric models, including neural networks and Gaussian Process, as the triggering kernel of Hawkes Point Processes.

Uncovering Social Links through Stochastic Point Processes [PDF][Code](TODO)

The combination of Hawkes Point Processes and Expectation Maximization Algorithm has been successfully applied to seismic model and social network model. To discover parenthoods between tweets in cascades, we borrow the same idea but strength it by using a recently-proprosed feature-driven generative triggering kernel. By comparing our method with NETINF and basic probability-based methods, we find our method can retrieve hidden parenthoods more accurately.

Face Detection and Recognition Using MTCNN and VGG Face Recognizor [PDF] [Code]

This work is for the project of Computer Vision (ENGN6528 @ ANU). Given three photos of all classmates taken from different directions, how shall we detect and recognize them? This thesis introduces an effective way which combines MTCNN and VGG face recognizor. MTCNN is used to detect all people in photos and VGG to identify detected people. Among three photos, one is used as calibration data to refine slightly parameters in the originally-trained model and other two are used to evaluate the performance of this method.