-
Randomized kD-trees
for searching/clustering high dimensional data
Finding nearest neighbors takes up most time during clustering many points with k-means. Due to the high dimensionality ordinary kD-trees fail to reduce the neighbor candidates. This approximation algorithm enables the research group to greatly speed up visual vocabulary construction and assignment.
Involved: C, Vectorization, OpenMP
-
Spectral clustering of image matching graphs
This implementation is used for comparison as a graph segmentation with good results.
Involved: Matlab
-
Extremely randomzed trees
for classification; won UCSD Data Mining Contest
I implemented the algorithm to be able to play with a lot of different ideas for modification.
Involved: Python
-
Distributed whiteboard application;
for Software Engineering lecture, team of 5
Involved: C++, Qt, RakNet
-
Open source projects contribution:
Python (bug fixes), Go (library feature implementation)