16-18 May 2011 Korea University Seoul Korea



IEEE International Workshop on Pattern Recognition in NeuroImaging



Call for Papers


Keynote Speakers

Technical Program


Paper Submissions




Conference Venue



Travel Information


Tutorial on Machine Learning for fMRI

Morning Session: May 16, 2011, 09:40am - 12:00pm
Title Introduction to machine learning for fMRI

tutorialFrancisco Pereira

Princeton University, USA

This session is designed for people who would like to learn about applying basic machine learning methods to fMRI data. I will cover essential concepts and what could be described as the "standard" workflow, as well as several practical issues which can have a large impact on results (or lack thereof).
  • Basic notions (classifiers, examples, train/test)
  • Classifiers and generalization (inductive biases, linear/nonlinear)
  • Statistical testing predictions (binomial null model, cross-validation, nonparametric tests)
  • Using classifiers for "localization" (in space/time/behaviour) (defining a feature space, voxel selection, determining feature influence on results)
Practical Session
  • Toolboxes (MVPA toolbox and PyMVPA, typical workflow)
  • Publically available dataset sources
  • Regularization and its effects
  • Voxel selection methods and voxel behaviour
  • Data and classifier example preprocessing, options and effects
  • Dangers, pitfalls and sanity checking (a.k.a. "what do reviewers look for?", double-dipping, train/test contamination, reproducibility)
  • Group-level conclusions
Afternoon Session: May 16, 2011, 1:30pm - 4:00pm
Title Advanced machine learning for fMRI

Francisco Pereira
Princeton University, USA

This session is designed to give a sense of the breadth of approaches that have resorted to machine learning methods, usually for purposes beyond determining whether information is present, such as validating models of mental processing. I'll cover work from recent years and attempt to provide a unified perspective, as well as an introduction to tools that should facilitate the implementation of novel analyses.
  • Using classifiers as detectors
  • Validating models and testing hypotheses
  • Reconstructing stimuli from fMRI data
  • Information mapping (representational similarity, searchlight classifiers)
Practical Session
  • Software packages for searchlight classifiers and similarity structure (Searchmight, Simitar, MVPA toolbox and PyMVPA implementations)
  • Software packages for (convex) optimization (CVX, optimize)
  • Demonstration with synthetic and real data
  • "Many ways to scan a cat": multiple analyses of a public dataset
These tentative contents subject to change at the whim of the presenter and heartfelt audience requests.
E-mail comments to
Last update: Mar. 9, 2011