Carnegie Mellon Machine Learning Lunch seminar

Learning Patterns of the Brain: Machine Learning Challenges of fMRI Analysis

author:Mark Palatucci, Robotics Institute, School of Computer Science, Carnegie Mellon University
published: Oct. 21, 2008, recorded: May 2008, views: 410

Turn off the lights

You might be experiencing some problems with Your Video player.

Slides

Slides
0:00	Learning Patterns of the Brain Machine Learning Challenges of fMRI Analysis
0:34	Neuro-semantics Research Team
0:46	Magnetic Resonance Imaging
1:24	Applications: fMRI (1)
1:57	Applications: fMRI (2)
2:24	Machine Learning Challenges
3:15	Challenge 1: Building a Cognitive State Classifier
3:20	Cognitive State Classification (1)
3:40	Cognitive State Classification (2)
4:07	Cognitive State Classification (3)
4:51	Cognitive State Classification (4)
5:54	Related Work: Rustandi [2007]
8:16	Hierarchical Bayesian Modeling
9:36	Related Work: Niculescu [2005]
10:57	Parameter Sharing
12:01	Feature Sharing Classifier (1)
13:25	Feature Sharing Classifier (2)
14:58	Experimental Results: fMRI Classification (1)
15:45	Experimental Results: fMRI Classification (2)
16:35	Experimental Results: fMRI Classification (3)
17:20	Challenge 1: Take Away Points
17:56	Challenge 2: Feature Selection Criteria for fMRI
18:12	Feature Selection in fMRI
19:33	Feature Selection: Histogram of Accuracies
21:06	Feature Selection: Highest Chance Accuracy (1)
21:18	Feature Selection: Highest Chance Accuracy (2)
21:53	Feature Selection: Example
23:00	Graph of Highest (Expected) Chance Accuracies
23:39	Feature Selection: Multiplicity Gap (1)
24:55	Feature Selection: Multiplicity Gap (2)
25:08	Feature Selection: Multiplicity Gap (1)
25:29	Feature Selection: Multiplicity Gap (2)
27:00	Feature Selection: Experimental Results
28:30	Feature Selection: Multiplicity Gap (2)
28:39	Feature Selection: Experimental Results
29:28	Challenge 2: Take Away Points
30:03	Challenge 3: A Generative Theory of Neuro-activation
30:50	Generative Theory
31:44	Approach: Integrate corpus data and fMRI data
32:42	Semantic Features
33:41	Predicting Activation: Simple Linear Model (1)
34:41	Predicting Activation: Simple Linear Model (2)
35:32	Evaluating the Theory
36:13	What are the best set of semantic features? (1)
36:34	What are the best set of semantic features? (2)
38:33	Future Work...
39:19	Challenge 3: Take Away Points
39:52	Summary of Challenges
40:22	Thanks!
42:15	- Questions
43:07	- Questions
44:02	- Questions
44:48	- Questions
45:12	- Questions
47:29	- Questions
48:05	- Questions

Report a problem or upload files

If you have found a problem with this lecture or would like to send us extra material, articles, exercises, etc., please use our ticket system to describe your request and upload the data.
Enter your e-mail into the 'Cc' field, and we will keep you updated with your request's status.

Lecture popularity: You need to login to cast your vote.

Description

Functional Magnetic Resonance Imaging (fMRI) has given neuroscientists and cognitive psychologists incredible power to analyze the deep mysteries of the human brain. With this powerful imaging technology, however, many new challenges have arisen for the statistics and machine learning communities. In this talk, I will present an overview of fMRI and some of the current machine learning challenges. I will discuss recent work on hierarchical Bayesian methods for dealing with high dimensional, sparse data. I will also discuss the application of classical order statistics to the problem of feature selection. Finally, I will show some of our latest results combining a large text corpus with fMRI to produce a generative model of neuro-activation for arbitrary words in the English language.