Generative Models With Domain Knowledge for Weakly Supervised Clustering

Explore a comprehensive lecture on incorporating domain knowledge in deep generative models for weakly supervised clustering, with applications to survival data. Delve into Laura Manduchi's research on integrating pairwise constraints and survival data into clustering algorithms, enabling exploratory analysis of complex biomedical datasets. Learn about the challenges of unsupervised clustering and the importance of guiding algorithms towards desirable configurations using prior information. Discover how leveraging side information in biomedical datasets can lead to medically meaningful findings. Examine topics such as weekly supervised clustering, survival clustering examples, synthetic and real-world experiments, nonparametric priors, and conditional Gaussian mixtures. Gain insights into the application of these techniques to infant echocardiograms and other clinical variables.

Introduction
Overview
Weekly supervised clustering
Survival clustering
Survival clustering example
Generality process
Answer question
Synthetic experiments
Realworld experiments
How to define the number of clusters
Tradeoff
Nonparametric Prior
Survival Distribution
Results
Clinical variables
Summary
Second work
Strain clustering
Conditional Gaussian Mixture
Generality model
Optimization
Datasets
Noise
Infant echocardiogram
Conclusions
Questions