Tensor Factorization for Biomedical Representation Learning

Explore tensor factorization techniques for biomedical representation learning in this seminar by Joyce Ho from Emory University. Discover how tensors can capture patient representations from electronic health records, addressing challenges of missing and time-varying measurements while outperforming deep learning models in predictive power. Learn about the application of tensor factorization in learning node embeddings for dynamic and heterogeneous graphs, and its use in automating systematic reviews. Gain insights into novel machine learning algorithms addressing healthcare problems, including patient subgroup identification, data integration, and handling conflicting expert annotations. Delve into topics such as matrix factorization, CPT decomposition, computational phenotypes, bibliographic networks, graph neural networks, and Tucker decomposition. Understand the challenges and theorems related to tensor factorization, and explore its potential in various biomedical applications.

Introduction
Joyce Ho
Representation Learning
Matrix Factorization
Singular Value Decomposition
Nonnegative Matrix Factorization
Subgroups
Natural Extension
CPT Decomposition
Data Sources
Computational Phenotypes
Representation
Time
Interpretation
Error
Repair
Summary
Motivation
Automating
Citations
bibliographic networks
graph neural networks
Tucker decomposition
Community
Community Graph
Results
Recap
Funding
Questions
Cost
Existence and Uniqueness
Theorems
Challenges
Question