Projects
Risk360: An Auto-Evaluator of Your Risk Score Model
This is an evaluation framework for the risk score models in healthcare. More details of this ongoing work are coming soon.
Fair Collaborative Learning (FairCL) for Model Personalizatio
This general framework is designed to reduce individual disparities in model personalization and integrate a variety of fairness concepts. We provide a self-adaptive algorithm to learn
Breakthrough T1D (Formerly JDRF): A Multi-task Learning Framework for Developing Genetic Risk Score (GRS) Models of T1D for Multi-Ethnic Population
In this project, we develop multi-task learning methods to reduce the disparities among different ethnoracial groups for genetic risk score (GRS) models of Type-1 diabetes.
Modeling User Choice Behavior under Data Corruption with Robust LDT
The emergence of many new mobile Apps and usercentered systems that interact with users by offering choices with rewards. However, real-world user data is often prone to data corruptions. To handle this issue, we propose RobustLDT, a robust learning algorithm for the Latent Decision Threshold (LDT) model, together with a user screening algorithm for the detection of bad actors with teasing or anomalous behaviors.
DARPA: Context-Aware Biomarker Discovery and Health Monitoring by Adaptive Integration of Heterogeneous Smartphone Signals
This project aims to discover biomarkers in different contexts and understand user activities based on data collected from various sensors with an healthcare monitoring app.