Ye Chen Email & Phone Number

Seattle, Wa, Us

Develop machine learning models for frequency management for advertising applications.

Global, Cn

[ Tech Lead for the carpool pricing algorithms ]• Drove the adoption of deep learning for the eyeball conversion model for price optimization. Improved the gross profit rate by 2pp when using the best Origin-Destination-Time (ODT) price generated the new conversion model. Deployed to 30+ cities in Brazil. - Developed a wide & deep neural network with a model calibration layer to predict if a passenger will make a carpool request. - Developed spatial and temporal clustering algorithms based on embeddings from the conversion model to better define price optimization units.• Developed or guided team members to develop various components / algorithms, e.g., hyperparameter search, user segmentation, feature engineering, order completion rate prediction, etc.• Unified the simulators for various carpool pricing products (ODT pricing, subscription, long trip pricing, coupon). Reduced the cost of pipeline maintenance by 75%.Collaborated with product, operations and engineering on defining roadmaps, sharing findings, scheduling experiments, launching products, and so on.[ Didi's Local News App - developed content understanding algorithms]The Local News App was developed for Didi's drivers and passengers during the pandemic when carpool business was temporarily shut down in the International markets.• Developed a logistic regression model to classify news into categories (sports, politics, travel, etc). Enabled the “Category Tab” product feature. Improved user retention rate by 5pp.• Developed a model to determine if a news is a local news. Enabled the “Local News Tab” product feature. Improved CTR by 6%.

San Francisco, California, Us

• Developed various machine learning models for real-time and week-ahead demand / supply forecast. Collaborated with downstream teams to optimize matching, surge multiplier and incentives using the forecasts, and therefore shape the demand and supply.Real-time forecasting models• Drove DS and Eng efforts to develop and deliver Uber’s first real-time event-aware demand forecasting model from model ideation to the global deployment in 121 cities. Improved the accuracy by 25% during events (sports games, concerts and festivals). Alleviated undersupply issue during events;• Collaborated with engineers to prototype and implement 5 improvements to the baseline real-time demand forecast to address impactful model accuracy gaps. Improved model accuracy during non-events time by 17%. Deployed to ~100 cities. Designed a switch-back experiment and measured a 1% increase in the number of trips;• Prototyped models for airport real-time supply forecast. Enabled us to support 2.5x more airports due to improved model performance.Week-ahead forecasting models• Developed a week-ahead holiday model that improves accuracy by 31%. The reduction in the forecasting error led to 2.7% (absolute) less deviation between budget and actual for a driver incentive program;• Developed and executed on a roadmap to improve the accuracy and spatial granularity of the week-ahead prediction model for marketplace balance. It enabled repositioning drivers at sub-city granularity and more efficient use of incentive budget for ~100 cities.Pipeline / automation• Automated data/model pipelines, including holiday model training, event model quality evaluation, and event metadata cleaning. Reduced manual effort by 80%.Experimental design• Designed and analyzed switchback and synthetic control experiments to evaluate the business impact of real time forecasting models.

St. Louis, Missouri, Us

• Lead seeding rate recommendation research program;• Discover, explore and obtain both internal and external data sources, evaluate their potential use in seeding rate research;• Develop and improve seeding rate recommendation models.

St. Louis, Missouri, Us

• Develop prediction models (regularized regression, random forest, etc.) to predict corn yield;• Apply clustering algorithms (k-means, hierarchical, model-based, etc.) to identify sub-field zones;• Develop and improve matrix factorization methods for corn seed recommendation;• Lead research projects;• Manage hiring process for two senior researcher positions for our team.

Armonk, New York, Ny, Us

Develop machine learning and data mining algorithms for two projects in electric energy industry.

Us

• Performed clustering analysis (k-means, hierarchical) to identify important features from brain images.• Predicted disease status using support vector machine and logistic regression.• Analyzed brain images (10GB scale) from our collaborators using the methods we developed.• Developed algorithms to learn fuzzy cognitive maps (FCM, a combination of neural nets and fuzzy systems)• Analyzed relation between genes from time series using clustering, correlation, Bayesian networks and FCM.• Supervising a rotation student’s research on brain dynamics simulation (10/2013 – 04/2014).• Drafted grant proposals with Dr. Lu (PI) in image analysis (local image features, deep learning).• Published 5 peer-reviewed articles, 1 book chapter, and 2 more articles in preparation.

• Developed power system security assessment algorithms based on kNN, decision trees and SVM.• Deployed a security assessment module to Sichuan Electric Power Corp., which serves 100 million people.• Developed evolutionary algorithms (GA, ACO and PSO) to optimize power generation cost.• Published 10 peer-reviewed articles.

Doctor Of Philosophy (Phd), Electrical Engineering

Master Of Science (M.S.), Power System Engineering

Bachelor Of Arts (B.A.), Electrical Engineering And Automation