I am Weiran Huang. Currently, I am a fourth-year computer science undergrad at Beijing University of Posts and Telecommunications. Previously, I was an R&D intern at MOMO Inc., a research Assistant at Chinese Academy of Sciences and Penn State University respectively. In Feb. 2020, I will start a new internship at Microsoft Research Asia.
In my opinion, computer science is a field where scientific research and engineering are closely combined. As one who has experience in both academia and industry, my career ambition is to bridge the gap between advanced theories and real-world applications to make humans’ life better.
To achieve this goal, my current plan is to pursue a master’s degree in CS or ECE in North America starting from Fall 2020.
I will join the ARD Incubation Group at MSRA working on Spatial-Temporal Prediction problems for traffic scheduling and resource optimization from Feb. 11th 2020.
I worked in the 3D Reconstruction Group, Deep Learning Lab at MOMO Inc., under Dr. Tianxiang Zheng’s mentorship.
- Developed a real-time face tracking tool which introduced dynamic rigidity prior to 3D face reconstruction, increased the stability score by 25.7% under drastic poses and expressions.
- Translated an existing MATLAB project into C++ using OpenCV, and modularized the code for easier maintenance and higher efficiency.
- Replaced the projection method in our face-changing app, “ZAO” with perspective projection, increased the reconstruction accuracy by 11.4%.
I researched on graph adversarial learning, speficically, robust graph neural network models, under the supervision of Prof. Suhang Wang at Penn State University.
- Extensively studied several academic papers published in top conferences on graph adversarial learning.
- Implemented several graph adversarial attack algorithms in Python, achieved comparable results as reported in papers but lower overhead.
I researched on network embedding and graph neural networks under the supervision of Dr. Shu Wu at CASIA. I was also a member of Scientific Innovation Training Program.
- Actively participated in several research projects on recommender systems and graph mining.
- Proposed a novel Graph Convolutional Network variant GraphAIR with cooperators, which is the first to explicitly take into account the non-linear neighborhood interactions. The paper of GraphAIR has been submitted to TKDE, preprint available here on arxiv.
- Implemented GraphAIR in Tensorflow, and it outperformed all baselines by significant margins on node classification and link prediction tasks, ranking first, second and fourth on 3 benchmark datasets respectively on paperswithcode.com.
Some of the following projects are my individual projects, while others are for my on-campus courses. Quarto is the project I am actively working on recently.
The following paper was finished while I was a research intern at NLPR, CASIA.