Modeling Electronic Health Records

Dynamic Graph-based Positive-unlabeled Learning Framework for Modeling Electronic Health Records

science formula and math equation abstract background. concept of machine learning and artificial
By Medical & Imaging Informatics (MII)
1 min read

UCLA will develop a dynamic graph-based positive-unlabeled (PU) learning approach for Electronic Health Record (EHR) analysis. This project focuses on advancing methods for leveraging complex EHR data. Key contributions of this research include:

  • Developing a novel dynamic graph learning framework specifically designed for EHR analysis, which is capable of modeling multi-granularity temporal patterns within an integrative context.
  • Implementing a structure-aware negative sampling strategy for the effective training of dynamic graph models in a PU-learning setting.

This work aims to enhance the analytical capabilities for understanding and utilizing the vast amounts of information present in electronic health records.

Contact PI: Corey Arnold

Funding Source: Optum Labs Topaz, Inc.