Jun 2020 – Apr 2023 · 2 yrs 10 mos · Mountain View, California, United States

May 2019 – Aug 2019 · 3 mos · Greater New York City Area

• 1. Performed tree-lasso guided high dimensional multi-response regression to find driving instruments for groups with 100k instruments using only 1-year training data
• 2. The R-Squared scores between the production book-level 10-year VPLs and that computed from the selected instrument sets are higher than 0.99; Overall selection ratio is 51%
• 3. Already put into production and proved to be successful on multiple lines of business

Jun 2018 – Aug 2018 · 2 mos · Beijing City, China

• 1. Conducted semi-supervised learning to generate training sets for labeling the users with interested categories with tens of millions of tweets in hot lists and a small golden truth list.
• 2. Filtered abnormal users with deliberately used wrong hashtags at the end of their tweets.
• 3. Extracted features to help replace the previous manually life cycle decision systems of pushed hot tweets with machine learning based ones.

Oct 2017 – Jan 2018 · 3 mos · Beijing City, China

• 1. Statistically analyzed major characters influenced the performance of more than 1,300 Didi Partners across China and designed Certification Standards for them.
• 2. Performed clustering and outlier detection algorithms to analyze abnormal increasing of drivers.

Jun 2017 – Sep 2017 · 3 mos · Beijing City, China

• 1. Performed imbalanced and supervised machine learning modeling on customer complaints with about 1.5 million records of orders and only 0.02% complaint rate.
• 2. Synthesized random under sampling, SMOTE, easy ensemble and random forest methods to build efficient, robust and explainable model after multi-dimensional attempts; final AUC is 0.80.
• 3. Effectively explained the model and provided detailed suggestions to the operation team, which won appreciation from all operation managers related.

Apr 2017 – Oct 2017 · 6 mos · Beijing City, China

• Publication: Wang, S., Li, Y. F., & Wang, Y. K. (2018). Hybrid Uncertainty Model for Multi-State Systems and Linear Programming based Approximations for Reliability Assessment. IISE Transactions, DOI: 10.1080/24725854.2018.1468123.
• Adjusted dynamic reliability approximation for reliability bounds of multi-state systems with hybrid uncertainty via dual relaxation, achieving relative error less than 1e-13.
• Pointed out the inaccuracies of results for classical three-element transmission system, rewrote the programs and redesigned the case study to show clear convergence of the algorithm and extend its application to far more complicated systems.
• Conducted systematical experiments to compare our methods to nonlinear approaches to demonstrate the advantage of the Approximation Algorithm.