Here is a PDF version: —->CV

Education

• Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences(SICA), 01/2019-06/2019
  • Visiting Student
• B.S. in Southwest Jiaotong University, Chengdu, Sichuan, 2019 (expected)
  • Major in Computer Science and Technology. 09/2016 - Now
  • Major in Pharmaceutics Engineering. 09/2015 - 07/2016

Related Experience

• 08/2018: Investigation on the Kaggle Competition Home Credit Default Risk based on Complex Modelling Analysis details
  • Exploratory Data Processing and Feature Capture
    • Checked missing values and visualized data distribution pattern; Performed imputation through replacing certain positive and negative INF values with NAN for Neutral Network (NN).
    • Followed the former preprocessing to do data feature extraction; Facilitated the utilization of feature variables concerning one hot encoding. • Applied LightGBM model to refine the summarized features regarding their contributions to the model.
  • Feature Engineering • Original Features: Obtained new features based on data computing integration.
    • Statistical Features: Enhanced the numerical variables given the categorical variables to summarize statistics.
    • Timing characteristics: Fixed time window and number of times to construct new features and combined summarized statistics.
  • Single Model Optimization
    • Dielectric argumentation on applicable models.
    The inborn missing data in the data prediction negatively affected the accuracy of Neutral Network (NN);
    Catboost was promising for dataset containing multiple variables, but the running duration was too long.
    LightGBM and XGBoost could possibly accommodate the prediction due to they can automatically impute the missing values regarding the reduction on training loss during the learning process.
    • Performed Bayesian Optimization to extend the optimal combination of hyper-parameters
  • Stacking Model Optimization
    • Performed 10 fold cross-validation to obtain out-of-fold prediction values as training and testing data on Logistic Regression model.
    • Strengthened the data optimization based on the stacking model integrating LightGBM and XGBoost model.
  • Results:
    • The optimized results indicated that the Area Under Curve (AUC) evaluation on predicting performance reached 0.7979; Achieved a brown medal.
    • The abilities of loan repayment depended on the previous applications requested to the credit bureau.
• 04/2018: Research on Internet Search Engine via Two Dimensions: Automatic Downloading and Text Classification details
  Supervisor: Prof. Xiao Wu, NLP, ML.
  • PART ONE: Automatic Downloading Process for English & Chinese Webpages
    • Applied Beautiful Soup Database in Python to extract and condense the content of HTML webpages on Jupyter and then selected the useful information.
    • Deleted the Stop Words in English via Natural Language Toolkit (NLTK) to facilitate the Porter Stemming and saved the output of simplified TXT.
    • Performed Porter Stemming for Stop Words in Chinese in the similar way as those in English and saved the output of simplified TXT.

  • Achievements:
    • All the steps were successfully integrated into a complete automatic system.
    • Extended the chart for common Stop Words to generate adaptable chart for specific Stop Words.

  • PART TWO: Text Classification Based on the Similarity of Different Texts
    • Mechanism for Similarity Comparisons:
    Set the vector of the words frequency given the times of word appearance.
    Calculated the similarity of words regarding the cosine distance of their vectors.
    • Mechanism for Text Classification:
    Transferred the text to matrix of word frequency through Term Frequency-Inverse Document Frequency (TF-IDF).
    Performed class analysis via K-means algorithms and optimized the value of k based on Elbow Rule to divide the data into two categories.
• 05/2017-03/2018: Exploration on Overlapping Community Detection and Its Application on Graph Data Mining details
  Supervisor: Prof. Hongmei Chen, Graph Data Mining.
  • Preprocessed network dataset by extracting data information file into node and edge files.
  • Applied small data clusters from Stanford Large Network Dataset (SNAP) to do visualized community structure analysis for Label-based Propagation algorithms (LPA), Clique Percolation Method algorithms (CPM) verification with Gephi.
  • Optimized CPM with Weak Clique Percolation Method(WCPM) replacement to decrease the time complexity of the algorithm through transmitting exponential relationship to linear relationship.
  • Improved a Label Propagation Algorithm for better mining performance and application on overlapping community.

Honors and Awards

• Students Scholarship for five times. 2016 - 2018
• Excellent volunteer in Life Mystery Museum at Chengdu city. Mar. 2016
• Third prize, China Undergraduate Mathematical Contest in Modeling. May. 2016

Skills

• Programming Languages: Python, R, Java, C, C++, Scala.
• Python Packages: Pandas, Numpy, Matplotlib, Scipy, Sklearn, Beautifulsoup, Seaborn, NLTK, Tensorflow, Keras.
• Software & Tools: Jupyter notebook, Gephi, LaTeX, Spark.

Extracurricular Activies

• A member of debating team in school of Life Science.
• Volunteer in Life Mystery Museum at Chengdu city.
• A member of Mathematical Modeling Institute in Southwest Jiaotong University