With my adviors on Princeton day of Optimization.
I participated the poster session in Princeton day of Optimization and present my recent work about Optimal Nonlinear Trees for Regression.
Biography
I am a 3rd year Ph.D. candidate at the MIT Operations Research Center supervised by Prof. Dimitris Bertsimas. My primary research interests are at the intersection between machine learning and optimization, with applications to healthcare and business.
I am currently working on the algorithm about interpretable machine learning method like decision trees. We use global optimization solution to replace the original greedy method when building decision trees in order to increase the accuracy of the whole model. Using the tool that we developed, we improve the prediction rules of children after head trauma who need computed tomography (CT). We also propose a new rule for liver allocation using the advanced machine learning and optimization tools we developed.
Before coming to the MIT, I earned my BS in mathematics with Summa Cum Laude from Peking University, China.
My email address is yuchenw "at" mit.edu.
Research
Interpretable Machine learning Method
"Optimal Nonlinear Trees for Predictions" with D. Bertsimas and J. Dunn, submitted to Machine Learning.
"Improving optimal trees using neural network" with D.Bertsimas and M.Sobiesk, working paper.
"Open Black Box Algorithm via Optimal Trees" with D. Bertsimas, working paper.
"Development and validation of an Optimized Prediction of Mortality (OPOM) for candidates awaiting liver transplantation" with D. Bertsimas, J. Kung, N. Trichakis, R. Hirose, P. Vagefi, 2018, American Journal of Transplantation.
" Building on Success: Improving the PECARN Head Trauma Rules With Optimal Classification Trees" with D. Bertsimas, J. Dunn, S.Dale, T. Trikalinos, submitted to JAMA Pediatrics.
"Visualizing tradeoffs in liver transplantation: the benefits of a continuous distribution model and its implications for national policy development " with D. Bertsimas, T. Papalexopoulos, N. Trichakis and P. Vagefi, submitted to JAMA.
"Spatial Temporal Choice Modelling in Crude Oil Seaborne Trade Using Machine Learning" with Haiying Jia, Roar Adland, submitted to Transportation Science Part B.
Machine learninig Applications
Activities
Industry Experience(Research Assistantship)
Quest, US (2019)
- Build interpretable machine learning model to predict high-cost claimants and help to decrease the total cost of the clients.
TheOfficialBoard, France (2018)
- Trained a Recurrent Neural Network (RNN) model to predict the department of people based on his job title, used Global Vectors for Word Representation (GloVe) to make recommendations of short titles based on their long job titles.
ISNetworld, US (2017)
- Used advanced machine learning method to develop a recommendation system for the client.
Estrategia, Peru (2016)
- Helped the largest family fund of Peru to build a portfolio management model and implemented time-varying linear regression method for predictions.
Teaching Experience
- Teaching assistant for Ph.D. and Master Level course 15.095 Machine Learning under modern optimization lens at Sloan School of Management, Fall 2018
- Instructed 4 Masters of Business Analytics Students to work with McKinsey Analytics about Spatiotemporal Analysis of Industrial Agglomeration and extracting named topics from unlabeled text data
- Course manager for the Analytics Edge on EdX course offered to more than 20,000 registered students, Spring 2016.
Pictures
My picture on the first floor of Sloan School of Managment.
I was discussing with a MBAn student Eric Green about their capstone project.