Neighborhood and Community Aware Clustering

2017.09 - 2017.11
Advised by Prof. Shih-Fu Chang and Dr. Svebor Karaman.

  • Developed an unsupervised approach to find the parameter setting with highest supervised measure score for the proposed Neighborhood and Community Aware Clustering method, and therefore created an automatic version of the proposed clustering method which can automatically choose an nearly optimal parameter setting.
  • Researched and experimented on the different preferences of different clustering criteria. Found the one most similar to the supervised B-cubed F1 measure and modified it to be an unsupervised graph based clustering criterion which is highly correlated with the B-cubed F1 measure on the clustering results obtained by our proposed clustering method.
  • Work submitted to ECCV2018.

Teaching Dimension with Partial Knowledge to Learner Hypothesis Set

2017.10 - 2017.12
Advised by Prof. Daniel Hsu, collaborated with Qinyao He

  • Investigated the estimating of teaching dimension with restricted knowledge to learner’s hypothesis set using interactive learning.
  • Proposed a reasonable setting to formulate the problem, and proposed several improvements on the upper and lower bound of teaching dimension under this setting
  • Course project in seminar course: Introduction to Learning Theory

Real-time Illumination Robust Face Tracker

2017.06 - 2017.08
Advised by Prof. Chengbin Ma, collaborated with Huan Zheng, Shengjie Pan, Siying Li, Zhenren Lu

  • Developed a real-time face tracking system that is robust under variance illumination and occlusion by combining the Kernelized Correlation Filter for tracking and the Faster RCNN for face detection.
  • Github Repo

Movie Face Clustering

2016.09 - 2016.11
Advised by Prof. Jia Deng, collaborated with Mingzhe Wang.

  • Designed a way to extract positive (same) and negative (different) relations between face tracks in the movie only based on the cuts and scene threads information. The result can help improve the performance of the movie face clustering algorithm.
  • Implemented a web based well designed labeling system. With the labeling system, me and Mingzhe labeled over 50 movies and 30000 face tracks. The results provided the necessary data to train our model
  • Researched and implemented several clustering algorithms such as Hierarchical cluster, Xmeans cluster and Kmeans cluster, with several cluster number determination method such as the Bayesian Information Criterion and Elbow method. Provided the baseline results on the movie face clustering problem.
  • Contributed to the design of the new proposed movie face clustering algorithm, including the feature selections and clustering scheme

Quantum machine learning and quantum algorithms

2016.05 - 2016.08
Advised by Prof. Yaoyun Shi

  • Researched about several Quantum Machine Learning Algorithms such as Quantum Support Vector Machine and Quantum Kernel Method.
  • Researched and mastered the improvements have been made in the last six year on quantum algorithms that solve linear systems and Hamiltonian Simulation problem.