Clustering Interval-Censored Time-Series for Disease Phenotyping

Unsupervised learning is often used to uncover clusters in data. However, different kinds of noise may impede the discovery useful patterns from real-world time-series In this work, we focus on mitigating interference interval censoring task clustering for disease phenotyping. We develop a deep generative, continuous-time model data that while correcting censorship time. provide conditions under which and amount delayed entry be identified noiseless model. On synthetic data, demonstrate accurate, stable, interpretable results outperform several benchmarks. clinical datasets heart failure Parkinson's patients, study how can adversely affect Our corrects source error recovers known subtypes.