- This event has passed.
A Crash Course in Statistical Learning Methods, Part I
May 15, 2018 @ 12:00 pm - 4:00 pm
This course, led by Olga Demler, PhD and Franco Giulianini, PhD, will review the Machine Learning methods used in medical research. The material will be split over two days (see topic outlines below). Laptops are required. Please install the programs R and RStudio before the course. Please register as space is limited (HERE for Part 1, and HERE for Part 2).
- Become familiar with the intuition behind each method and the language used in the field
- Gain hands-on experience using these algorithms in the R programming language
- Working knowledge of intermediate statistical analysis including linear and logistic regressions and linear discriminant analysis
- To participate in the practice exercises (which are optional), beginner-level proficiency with R programming language is required (an equivalent of completing weeks 1, 2 and 3 of https://www.coursera.org/learn/r-programming — JHU “R programming” course on Coursera). Please bring a laptop with R and RStudio installed from www.r-project.org and www.rstudio.com.
This material is based on recent developments in the field (references will be provided) and the book by Friedman, Hastie, Tibshirani “The Elements of Statistical Learning” (http://statweb.stanford.edu/~tibs/ElemStatLearn).
TOPICS: Statistical Leaning Methods I (May 15th) – Register HERE
- Supervised Learning Algorithms
- Classification and Regression Trees
- Random Forests
- Support Vector Machines
- Feature Selection Algorithms
- Ridge Regression
- Elastic Net
- Deep Learning
- Convolutional Neural Networks
- Recursive Neural Networks
TOPICS: Statistical Leaning Methods II (September 27th) – Register HERE
- Unsupervised Methods
- Dimension Reduction
- PCA vs Factor analysis
- Classification and Pattern Recognition
- K means clustering
- K nearest neighbor classifier
- Dimension Reduction
- Short Review of Supervised Learning Methods: Elastic Net, Random Forest, SVM
- Other Topics:
- Multiple Comparison Techniques: controlling for family-wise type 1 error rate: Bonferroni, FDR, permutation-based FDR to adjust for correlated biomarkers