My name is Shengpu Tang (唐笙普), and I am a PhD candidate in the computer science department at the University of Michigan. I'm a member of the Machine Learning for Data-Driven Decisions (MLD3) research group led by Jenna Wiens.
My research focuses on developing and applying machine learning methods to help solve important problems related to healthcare. While motivated by specific use cases (such as risk stratification and dynamic treatment recommendation for various diseases and medical conditions), my work aims to identify interesting technical challenges and propose novel solutions that are broadly relevant to fundamental AI/ML research. My past contributions span the technical areas of reinforcement learning, deep learning, time-series/sequential data modeling, transfer learning, causality, algorithmic fairness, etc., and I hope to apply my knowledge and expertise to interdisciplinary problems in healthcare and beyond.
PS: I have many genius friends and this LINK will redirect you to one of them.
- Artificial Intelligence
- Machine Learning
- Healthcare
-
MS in Computer Science & Engineering, 2020
University of Michigan, Ann Arbor
-
BSE in Computer Science, 2018
University of Michigan, Ann Arbor
Media Coverage
- In scramble to respond to Covid-19, hospitals turned to models with high risk of bias. MedCityNews. April 21, 2021.
- ★ From collaboration to validation: How a team of engineers, clinicians, programmers, and administrators evaluated one of the most widely implemented predictive models. Precision Health at UM. March 30, 2021.
- Epic Deterioration Index May Identify High- and Low-Risk COVID-19 Patients. Clinical Advisor. January 22, 2021.
- New framework helps streamline EHR data extraction. Healthcare IT News. October 12, 2020.
- Faster than COVID: a computer model that predicts the disease’s next move. The Mighican Engineer News Center. May 12, 2020
- Computer scientists employ AI to help address COVID-19 challenges. Michigan AI Lab. April 15, 2020.
Featured Publications
Model Selection for Offline Reinforcement Learning: Practical Considerations for Healthcare Settings
We present an in-depth analysis of popular OPE methods when used for model selection in offline RL, highlighting the additional hyperparameters and computational requirements, and propose a simple two-stage approach to accelerate model selection while balancing accuracy with computational costs.
Democratizing EHR Analyses with FIDDLE - A Flexible Data-Driven Preprocessing Pipeline for Structured Clinical Data
We developed FIDDLE (Flexible Data-Driven Pipeline), an open-source framework that streamlines the preprocessing of data extracted from the EHR. We also released preprocessed datasets in the following publication:
MIMIC-III and eICU-CRD: Feature Representation by FIDDLE Preprocessing. PhysioNet. April 2021. https://physionet.org/content/mimic-eicu-fiddle-feature/1.0.0/
Clinician-in-the-loop Decision Making: Reinforcement Learning with Near-Optimal Set-Valued Policies
We proposed a model-free algorithm based on temporal difference learning and a near-greedy heuristic for action selection – with the goal of learning set-valued policies to capture near-equivalent actions that lead to similar cumulative rewards.
Recent Publications
Relaxed Parameter Sharing: Effectively Modeling Time-Varying Relationships in Clinical Time-Series
A BioHackathon on the challenges of aging: Rapid need-finding and solution modeling in a student-led event
Visitors
Activities
short films, games, etc.
Set Sail
Campus Movie Fest 2019
Set Sail (part 2)
Campus Movie Fest 2019
FreshWalk
EECS 598 - Advanced Data Mining
pandas
Contributed bug fixes
Legend of Zelda NES remastered
EECS 494 Game
Zelda Windy_World
EECS 494 Game
FlappyBird coop
EECS 494 Game
Flash-mob Oct.1 @ UM diag
Celebrating National Day of China, 2015