Bowei Wei

ac8nen@r.postjobfree.com j 774-***-**** j LinkedIn: bowei-wei-6316975a/ j Github: BoweiWei

** ******* ******* **, **********, RI, 02904

EDUCATION

Brown University Providence, RI

Master of Art in Biostatistics; GPA: 3.5/4.0 Expected May 2019 Relevant Coursework: Machine Learning, Deep Learning, Data Analysis, Data Mining & Text Mining, Introduction to Data Science, General Linear Model

Worcester Polytechnic Institute Worcester, MA

Bachelor of Science in Actuarial Science; GPA: 3.3/4.0 Aug 2012 - May 2016 SKILLS

Languages: R, Python, SAS, Easylanguage, Visual Basic Advanced, SQL, HTML, LaTeX and Excel Technologies: GCP, GitHub

Libraries: TensorFlow, Numpy, Pandas, BeautifulSoup, Jupyter, OpenCV, PIL, dplyr, ggplot2, knitr, RSQLite, shiny EXPERIENCE

Brown Kaggle Competition Providence, RI

Competition Participant (Machine Learning) September 2018 - December 2018

Data Prediction: A General Linear Model to Predict Upcoming Attendance of Brazil Hospital Appointment

Worked in a team of four people and came up with a model for predicting whether the patient will show up in the next appointment

Implemented multiple models including KNN, SVM, Neural network, random forest, XGboost and so on to find the best fit model for training data

Ensembled the predicted data for better log loss at around 67.854 which make us the 4th out of 40 teams. PROJECTS

Art Style Classification Using Alexnet: A Convolutional Neural Network For Graphic Classification Using Tensorflow in Python

Designed and implemented an shrinkage and randomly selected images from 60,000+ images to prepare a clean dataset of 28*28 clean images from the dataset for training.

Redesigned and trained an eight layers Alexnet with two modified layers to classify 35 styles of art resulting 50%+ accuracy with random selected images for 1 epoch.

Achieved a 80%+ top3 accuracy with shrinkage image after 15 epochs on Painter by Number dataset from Kaggle

Fake Face Manufacture Using Deep Learning: Deep Convolutional Generative Adversarial Network using Tensorflow in Python

Manufactured the faces of these virtual companions by implementing a Deep Convolutional Generative Adversarial Network (DCGAN), which uses a 10,000 images dataset consisting 64*64 Real Image Data of celebrity faces to generate new faces using tensorflow on Google Cloud Platform

Achieved FrÃl’chet Inception Distance(FID) around 460 in 6 epochs

Natural Word Advisor From Penn Treebank Corpus: Recurrent Neural Network Language Model Using Tensorflow in Python

Built a Recurrent Neural Network Language Model with Word Embedding for language modeling the Penn Treebank Corpus using Tensorflow with Windows size 20, batchsize 50.

Achieved the perplexity around 180 using AdamOptimizer and learning rate 1E-3.

Cartpole Self Balancing System: AI Algorithm For Self Balancing game Using Tensorflow in Python

Implemented the Advantage Actor-Critic (A2C) Algorithm for the Cartpole-v1 task in OpenAI Gym

Created two feed forward models, one for the actor and one for critic and used sum of actor loss and critic loss to update the actor and critic.

Achieved 400+ mean reward in the last 100 episodes after training 1000 episodes.

Modified the Actor Critic algorithm into a REINFORCE algorithm in the same model and achieved very same result. RELEVANT COURSES

Core Courses Other Courses

Deep Learning Methods in Epidemiological Research

Practical Data Analysis Probability and Statistical Inference Data Science with SQL Statistical Programming with R Applied General Linear Model Programming with Python Statistics Learning and Big Data Basic of Public Health

Contact this candidate