Sentimental Analysis based on Convolutional Neural Networks
- Utilized distance supervision approach to trained a 2-layer Convolutional Neural Networks for classifying positive sentiment and negative sentiment based on the 200M dataset from IMDB and Yelp.
- Created the word embedding for generating the sentence matrix model for each sentence.
- Trained the pre-processed word vector in the 2-layer neural network, and accuracy on the test set was 53.7%
Large Scale Patent Classification
- Classified large-scale patent with over 110,000 samples size using supervised learning model.
- Utilized different method to classify, including Naive Bayes Classifier, SVM and Max-Min Algorithm (include both Random-based and Prior Probability-based).
- Utilized Multiprocess Library from Python to achieve parallel execution and higher performance.
Efficient Parallel String Searching
- Implemented the multiprocessing programs for string searching algorithm using OpenMp and CUDA via GPU, based-on Naive algorithm, BMH (Boyer-Moore-Horspool) algorithm and KMP(Knuth-Morris-Pratt) algorithm.
- Proposed a novel method based-on BMH algorithm for string searching which is more suitable for parallel.
Tiger Compiler Implementation
- Added Lexical Analysis Phase to divide the Tiger source code in to tokens.
- Developed the Abstract Syntax Tree.
- Generated Intermediate Code (IR Tree).
- Translated the intermediate code into target machine language MIPS Assembly
Advanced MIPS CPU Simulator with Multi-Cycle/Pipeline
- Developed the simulator of each components used in a MIPS CPU like instruction memory, data memory, ALU and ALU Control on C++
- Connected all components into an advanced MIPS CPU simulator with multi-Cycle and pipeline
- Implemented the multi-cycle CPU Simulator on a Xilinx Experiment Board in Verilog HDL