• Investigate given a simulation setting, whether positive deWinite projection of hessian matrix through partial
projection is possible at numerical factorizations step of Cholesky decomposition instead of a total projection before solving. 

• Helped developing GPU version of previous project EGGS, which now has significant speedup comparing
to state of art methods or EGGS itself. In one of the application the performance of shape deformation of large

• Investigated multiple interval libraries on their behaviours when dealing different type of functions,
correctness, speed and consistency across multiple platform.
• Developed a benchmark for interval libraries: https://geometryprocessing.github.io/intervals

• Developed method that unrolls expression tree for sparse matrix arithmetic and just-in-time compile kernel for the expression and can be easily plugged into current applications.
• The kernel supports CPU parallelism and AVX instructions. Current benchmark shows 4X-12X speed improvement in Sparse Matrix Sparse Matrix multiplication (SPMM) comparing to Intel’s Math Kernel Library. In real world applications, using this technique can eliminate the bottleneck posed by sparse computations. 

• Used unsupervised learning to analyze the syntactic pattern of a word for building a new type of noun database.
• Optimized the data structure and pipeline and used parallel programming to make the project deployable on server machine.
• Introduced new analysis pattern hence increasing the total amount of information extracted in a single sentence.