On Nonlinear Dimensionality Reduction and Its Applications in Biomedical Image Analysis

Pre-FPO Presentation
Date
Aug 1, 2023, 1:00 pm2:30 pm
Location
192 Nassau St, St 1

Speaker

Details

Event Description

Neighbor embedding (NE) algorithms are unsupervised algorithms that identify groups of related data. They are regarded as nonlinear dimensionality reduction methods, as they allow visualization of high-dimensional data in low-dimensional space (typically 2-D). A popular NE algorithm is Uniform Manifold Approximation and Projection (UMAP). UMAP utilizes a k-nearest neighbor (k-NN) graph to establish a pairwise metric in a high-dimensional space, which is then used to align a lower-dimensional representation. In the first part of the talk, we analyze the UMAP algorithm improve out-of-sample embedding, explain the learning rate annelation scheme, and improve the consistency of embeddings when initialized randomly. In the second part of the talk, we apply UMAP to biomedical image datasets. In particular, we analyze chest x-rays and show that dimensionality reduction can discover 1) different phenotypes of COVID-19 response and 2) anomalies in image datasets. Unsupervised algorithms like UMAP find hidden patterns and relationships in biomedical image analysis which leads to new discoveries in disease classification and methods for data curation.

Adviser: Jason Fleischer