Deyao Kong Email & Phone Number

New York, Ny, Us

美国 纽约

New York, United States

Remote

• Time-Series Modelling: Designed and compared three time-series approaches to forecast revenue, including moving average, exponential smoothing, and ARIMA; researched Amazon’s DeepAR model to deal with multiple time series and predicted the distribution of the likelihoods of different future scenarios showcasing a set of prediction intervals

Beijing, China

• Neural Network: Implemented the Logistic Regression and the Random Forest models on credit card transaction dataset for fraud detection and achieved a 0.88 AUC-ROC score; enhanced the score to 0.95 by applying neural network models with Pytorch, testing different loss functions, and utilizing k-fold cross-validation.• Regression Models: Retrieved data from Yahoo Finance and utilized multiple regression model to predict stock opening prices; applied Ridge and Lasso regression models to solve overfitting problems and decreased RMSE by 25%; improved the accuracy by conducting polynomial regression and MLP neural network models.• Data Visualization: Developed web crawlers to collect video game sales data from steamdb.info, etc.; cleaned and preprocessed the data before storing it in the customized MongoDB database; applied linear regression and the SVM model to analyze the sales data; utilized seaborn, folium, and Tableau to visualize the trends; results were demonstrated to students for a better understanding of the concepts.• Decision Tree Model: Developed a decision tree classifier from scratch; utilized information entropy criterion for the best splitting result; applied the model to wines dataset and achieved a 0.97 accuracy for predicting wine types.

Beijing, China

• Machine Vision: Researched and analyzed the algorithm of a human facial identification model based on YOLO v3 (darknet); tested different loss functions and parameters to optimize the deep learning process of the neural network model and decreased the training cost by 5%; designed testing models and confusion matrices for validation; supported the supervisor in setting up the neural network model according to the demand of the front-end department

Beijing, China

• Deep Learning: Utilized Python with OpenCV to develop a human fingerprint identification convolutional neural network based on Pytorch framework for employee security check and achieved 90%+ accuracy; accelerated the learning process by testing and implementing different loss functions• Organized weekly meetings and led other trainees from the front end to discuss thepros and cons of different algorithms; Summarized meeting transcripts to generateanalysis reports to the manager

Master Of Science - Ms, Data Science

Coursework: Data Structures & Algorithms, Intermediate Statistical Methods, Data Mining, Database Systems

Bachelor Of Science - Bs, Mathematics

Concentration in Applied Mathematics