1. AutoML for time series prediction2. Auto content moderation, team lead

• Develop several algorithms in machine Learning platform based on Spark• Develop Deep learning platform based on Tensorflow, including some distributed optimizers• Develop OCR algorithms

• Divided the patients data into different groups based on age groups and gender and then in each group builtprediction models. The accuracy was improved by 10% compared with the original model without gender and agegroups• By adding weights based on the age distribution of diseases rather than age groups, the accuracy was againimproved by 6%. Then developed an Adaboost model with above model as base learner to predict diseases and theaccuracy was again improved by 5%• Developed a deep learning model to predict diseases based on symptoms and inspection indicators. The accuracycan be as high as 55% due to the limitation of the original data• Designed a new Post-Pruning algorithm in decision trees to develop a Q&A system, and visualized the decisiontrees using Python• Converted more than 1000G unstructured Electronic Medical Record data to structural data by using Hadoop andfiltered the error data• Defined several attributes and then segmented each symptom into these attributes by CRF++. Found more than130,000 pairs of synonym symptoms and 200,000 pairs of father-child symptoms based on those attributes• Inspired by F-score and association rules, used support and confidence to calculate the typical concomitantsymptoms of diseases• Coordinated Data Team’s work in agile development and preside daily work summary meeting