Mose Wintner

- Technical lead role.

- Led retrospective study demonstrating state-of-the-art machine learning prediction performance in predicting various cancer diagnoses from electronic medical records using XGBoost, SHAP, scikit-learn, pandas (0.768 AUC @ two-year prediction horizon).

- Led design and implementation of distributed, flexible, fully traceable software packages in Python for reproducible, fast, space-efficient data preparation and feature engineering from electronic medical records and insurance claims.

- Provisional patent filed for select technology from feature engineering pipeline.

- Mentored associate-level scientists in machine learning and programming practices.

- Various other investigations using pytorch, including generative modeling of synthetic data..

- Led various research projects involving pediatric ICU data, including

• Patient-ventilator asynchrony detection using bespoke convolutional + recurrent neural networks, spectrograms, and wavelets.
• Prediction of critical deterioration from hospital floor to PICU using LSTMs in Keras.

- Construction of new and improvement of existing internal software libraries for use by other vPICU data scientists.

- Communicating with clinicians and supervisors to iteratively define problems and refine data.

- Validating, visualizing, filtering, and exploring anomalies in inpatient electronic medical record data.

- Planning, preprocessing, feature engineering, modeling, evaluation, implementation, automation, and presentation

- Led definition and construction of end-to-end predictive modeling solutions from planning through implementation, including

• Predicting 12-month mortality of patients with advanced kidney disease using tree ensemble model and SHAP.
• Detailed statistical analysis of patient emails to adult primary care

- Founded and led informal group and monthly enterprise-wide meeting for over 70 advanced analytics professionals to share work, ideas, and knowledge and build a robust culture of collaboration.

- Presented and led discussions on safety and trustworthiness of artificial intelligence for healthcare and multiple imputation of missing data.

- Consulted in identification of areas of opportunity for artificial intelligence at Kaiser Permanente.

- Led discussions for 20-100 students twice weekly.

- Courses taught included time series, optimization, dynamic modeling, Matlab for neuroscientists, real analysis, and linear algebra.

- Managed schedules of tutors and myself 2011-14 for Friends of Euclid.

- Led advertising and expansion strategies

- Tutored 8th grade through graduate level mathematics, statistics, physics, test prep (SAT, ACT, ISEE), economics, chemistry for Friends of Euclid and numerous other companies.