University of Wisconsin-Milwaukee Machine Learning 2023

Note: This page is still under construction. Check back soon!

Below is an annotated agenda for the workshop. To prepare for the course, do the following steps:

1. Get Started with Matlab

Please watch this playlist for an introduction to Matlab and its basic commands; you may also want to go through the tutorials here. It is also recommended that you download version R2021a or later.

2. Install SPM12

Instructions for downloading and installing SPM can be found here.

3. Install The Decoding Toolbox

Click here to navigate to The Decoding Toolbox website; click on the link that says "click here to download TDT", and then click the download button when it takes you to a Google Drive site. Then, follow the instructions here to install and set up the toolbox.

Day 1: Introduction to SPM12 and Machine Learning

Agenda

(9:00am-10:00am) Review of Matlab
This will be a brief review of Matlab and its fundamentals; these will be necessary for using the CONN toolbox, as well as more advanced methods such as scripting analyses. This lecture will cover:

  • Basics of navigation

  • Variables and Structures

  • Paths and Functions

  • Control Structures

  • Checking installation of SPM12; creating template job and loading it from the Matlab terminal.

(10:00am-12:00pm) Introduction to SPM12 (Lecture & Practical)
This first practical will be a guided hands-on tutorial about how to open the SPM12 GUI and examine some of its features. We will review the following topics:

  • History of fMRI: Landmark studies, common issues, uses and future potential

  • Functional vs. Structural images

  • Basics of preprocessing: slice-timing correction, coregistration, normalization, and smoothing

  • Overview of the SPM12 GUI

  • Creating a new project and loading the data

  • Initial quality checks

(12:00pm-1:00pm) LUNCH BREAK

(1:00pm-2:00pm) Basics of Machine Learning (Lecture)
This session will provide an overview of MVPA, a popular multivariate tool for neuroimaging data.

  • Overview of Machine Learning and hyperplanes

  • Uses of MVPA

  • Experimental design considerations for studies that will use MVPA: Timing schemes, masks, and how to validate the results

(2:00pm-4:00pm) Introduction to The Decoding Toolbox (Lecture & Practical)
We begin our first practical session of machine learning by using The Decoding Toolbox, a popular software program for machine learning analysis.

  • Overview of The Decoding Toolbox

  • ROI analysis vs. Searchlight analysis: Pros and cons of each

  • Confusion Matrices

  • Replicating the results from Haxby et al. (2001)

  • Non-parametric testing for significance

(4:00pm-5:00pm) Individual Consultation Sessions

Day 2: Machine Learning Advanced Topics

Agenda

(9:00am-10:00am) Representational Similarity Analysis
Representational similarity analysis (RSA) exploits the correlation similarity structure of voxels, using different distance metrics to illustrate the representational distance of different conditions.

  • Measurement of both content and format of representations

  • How this is used across modalities

  • Editing The Decoding Toolbox template scripts for RSA

  • Analysis of sample dataset

(10:00am-11:00am) Introduction to Python
More advanced analyses such as Hyperalignment require Python. In this lecture, we will review the fundamentals of Python, including packages, methods, and how to use Jupyter notebooks.

(11:00am-12:00pm) Hyperalignment
Hyperalignment is a relatively new classification technique developed by Jim Haxby’s lab, which aligns subjects' brain data in a high-dimensional space of voxels/features.

  • Introduction to hyperalignment

  • Benefits of hyperalignment vs. traditional MVPA: Alignment of functional topographies

  • Hybrid hyperalignment: Combining task and functional connectivity profiles

  • Analysis of movie datasets: Classifying which part of a movie a subject was watching

  • Application to other datasets