Courses Detail Information

ECE6706J – Unsupervised Learning for Science

Instructors:

Credits: 2 Credits

Pre-requisites: Graduate Standing

Description:

Unsupervised learning algorithms are methods for transforming and finding structure in datasets without the benefit of labeled examples to guide them. Students will learn the theory behind unsupervised machine learning models as well as a plethora of optimization algorithms. Students will also learn how unsupervised machine learning is related to single cell multi-omics analytics and recent advancements in both unsupervised learning as well as their applications in single cell multi-omics analytics. Unsupervised learning and single cell multi-omics analytics is an essential part of modern bioinformatics education program. In single-cell multi-omics, whch has been branded “method of the year” by Nature Methods in 2013, 2018 and 2019, rarely are datasets curated. Together with the high-uncertainty, high-dimensional and heavily-heterogeneous nature of single-cell multi-omics data, unsupervised learning has been the primary tool in single cell multi-omics analytics.

Course Topics:

Clustering
K-means and K-medoid
Model based methods (GMM, DIMM, etc.)
Graph based (Louvein and Leiden)
Deep Learning Based Methods
Applications in scRNA-seq Analytics
Feature Selection
Filter Methods (Information Gain, MAD, Variance Threshold, Dispersion Ratio, etc.)
Wrapper methods (Forwad Selection, Backward Elimination, Bi-Directional Elimination, etc.)
Embedded methods (Regularization, Tree-Based Methods)
Applications in scRNA-seq Analytics
Bayesian Frameworks
Bayesian Modeling
Bayesian Estimation
Variational Bayes Inference
Applications in scRNA-seq Analytics
Dimension Reduction and Manifold Learning
PCA
CCA
NMF
T-SNE
UMAP
VAE
Trajectory Analysis
Applications in scRNA-seq Analytic