Data Software Engineer

Chenwu Liao

******@*****.*** · 734-***-**** · ***0 Broadway Street ·Ann Arbor, MI 48105

EDUCATION

University of Michigan Ann Arbor, MI

Master of Science in Electrical and Computer Engineering May 2020 Major Area: Signal and Image Processing and Machine Learning GPA 3.88/4.00

Coursework: Foundation of Computer Vision, NLP Algorithms and People, Digital Integrated Tech, Computational Data Science, Machine Learning, Optimization Methods in Image Processing, Probability and Random Process, Estimation Filtering and Detection Nanjing University Nanjing, China

Bachelor of Science in Physics June 2018

GPA 3.59/4.00

EXPERIENCE

University of Michigan Center for Digital Infrastructure Finance Ann Arbor, MI Graduate Research Assistant June 2019-present

● Write Python script to retrieve S&P 500 companies’ Financial and Environmental data from Bloomberg API

● Develop learning algorithms using Neural Networks, Random Forest for a new index --- WaterBeta, convert original water intensity to quarterly based data by new learning algorithm

● Identify water Intensity index can be explained by financial features with r2 score 0.53

● Provide 10 years water intensity prediction data to Equarius Risk Analytics LimnoTech Ann Arbor, MI

Software Engineer Oct 2020-present

● Write python script to retrieve state level US census data from API using request

● redevelop ESG states score by low rank matrix completion to refill the blank data from Bloomberg using pandas and scikit learn Python library

PROJECT EXPERIENCE

Deep Learning models for Efficient Detection of Manipulated Images

● Created a deep neural network as a binary classifier trained on vast amounts of data(Resnet18 as Convolution Blocks) to get accuracy of 71.8% in validation and 64.2% in test

● Implemented Siamese Networks to determine self-consistency between patches of an image to identify inconsistent image-patches / splices to get validation accuracy of 72.6% and test accuracy of 64.9%

● Combined both trained models and work towards making a unified prediction from the outputs of both of these networks and boost the network validation accuracy of 73.8% and test accuracy to 66% Kaggle challenge Malaria Parasite Classification in Thin Blood Smear Images using Deep Convolutional Neural Networks

● Implemented baseline(SVM) and Deep convolutional Neural Network ML algorithms (AlexNet, VGG-16, Resnet50, Xception) to classify the malaria cells and infected blood cells

● Solved the high imbalance property of Medical Image data by using data augmentation and other types of loss function to get Accuracy of 98.61%

Design Yelp-like camping website

● used HTML,CSS Bootstrap CDN, Fontawesome,jQuery for frontend design

● Implemented comment system by seeding the database byNodeJS MongoDB for backend Propaganda detection of News Articles by NLP Deep Learning Models BERT and XLNet

● Implemented customized BERT(freezing layers, changing fully connected classifier) and XLNet to detect 14 different propaganda techniques in a corpus made by 550 news articles

● Achieved F1 score of 0.64 compared with baseline SVM’s F1 score –0.32 SKILLS

MySQL, Python,C++, HTML,CSS, Javascript,Node.JS, Java, MongoDB

Contact this candidate