Jun Wang

University of Michigan

Master's degree, electrical and computer engineering, machine learning and signal/image processing

January 1, 2015 – January 1, 2017

ServiceNow

Senior Machine Learning Engineer

September 1, 2019 – March 1, 2022

ServiceNow

Machine Learning Engineer

June 1, 2017 – August 1, 2019

Recommender System on Non-Metric Space (C++, Python)

January 1, 2019 – Present

Recommender System on Non-Metric Space

Golang Search Engine (Go)

January 1, 2017 – Present

• Designed 5 layers for a search engine by Go, including web, engine, index, segment and field layer. • Implemented inverted index, dictionary and key by B+ tree, accelerated data reading and writing by calling MMAP, sorted documents by semantic relativity evaluated by TD-IDF. • The search engine's QPS reached 1000 in a 4-core, 8GB RAM computer.

Artificial Intelligence Game Design (Python and C++)

October 1, 2016 – Present

• Designed the strategies for playing the tower defense game in a maze in Python; implemented DFS, BFS and UCS algorithm to find the goal under uninformed search; designed heuristic function for informed search, including Greedy Best-First and A* search, to choose the best solution. • Designed Snake game based on AI algorithms in C++; implemented two AI methods: Hamiltonian Cycle and Graph Search, realized the perfect snake filling out the map, where an approximate solution to NP-hard problem: the longest path was also implemented.

Predictive Model for Power Load based on Genetic Algorithm (R)

August 1, 2016 – Present

• Performed exploratory data analysis and predicted the power load in the whole day by the weather and date type (holidays or not, workdays or not). • Imputed missing values and smoothed raw data by Nearest Neighbors method; explored the pattern on time series by covariance and correlation analysis and visualizations. • Implemented error function in the latent layer, built and trained the Radial basis function(RBF) network on the regularized data, predicted power load at all the 96 time points in the whole days in a month by the optimized parameters.

Predictive Model for Financial Time Series (R and Excel)

June 1, 2016 – Present

• Researched and modeled the relation between future markets and macroeconomy. • Transferred raw time series into volatility and liquidity of major futures to evaluate price discovery. • Built GARCH model to evaluate the macroeconomic uncertainty and volatility • Performed ANOVA and t-test to select features, built multiple regression mode • Diagnosed model by residuals and cases, back-tested and optimized model, which could predict price discovery of futures under 90% confidence level.

Retinal blood vessels segmentation detection and seeded region growing (MATLAB and Python)

April 1, 2016 – Present

• Designed three steps for vessel segmentation, including image normalization and thin vessel enhancement, automatic seed selection by filters and segment validation, image region growing in multi-scale and segmentation enhancement. • Evaluated basic region-growing algorithms versus SRG method and achieved better performance in terms of average 91.3% accuracy, realized high-precision vessels reconstruction.

Machine Learning Projects (Python, R and Matlab)

March 1, 2016 – Present

• Implemented basic machine learning algorithms from theory to practice by Python and MATLAB, including SVM, linear regression, PCA, k-Means, EM, Naïve Bayes and etc.; • Packages in Python(scikit-learn), R(caret,e1071,randomForest,etc.), MATLAB(Statistics and Machine Learning Toolbox) were practiced in Hand-written Digit Denoising (PCA, Boost) and Recognition(Random Forest)(95% accuracy), Spam Filtering(Naïve Bayes)(92% accuracy) where the machine learning models were trained, tested and optimized.

Visual Object Tracking System (C++)

January 1, 2015 – Present

• Designed a new visual object tracking algorithm which combined Compressive Sensing and Particle Filtering. • Compressive Sensing was improved to compress visual objects and Particle Filtering, a Monte-Carlo method, was designed as the recursive tracking framework; implemented the algorithm by C++ • Achieved the average 92.1% accuracy of visual tracking, realized the real-time tracking in complicated scene.