Brain electrical grids and computing grids:A perfect match

Mike X Cohen

The brain

GREEN

The behavior

Brain electrical activity

Cutting-edge EEG recording technology

256 electrodes1 kHz sampling

~5 GB/hour

Mag

nitu

de

(a)

Phas

e an

gle

(q)

real

*real

imag.m

q

a

Data dimension explosion

21437174

Beamforming for source localization

Data dimensions:

1) Space (spherical or 3D)2) Time3) Frequency (power vs. phase)4) Experiment condition5) Connectivity6) Subjects

Why we need computational power

Example of typical results

It gets worse:Anatomical connectivity is also important.

Voxels

Voxe

ls

EEG connectivity matrix

Voxels

Voxe

ls

DTI connectivity matrix

EEG single-trial power MNE conflict effect Seeded tractography

Linking anatomical and functional connectivity

The problem:

Brain data analyses involve many repeated transformations of very large

data matrices.

Standard computers cannot handle these analyses without annoying interventions.

The solution:

A scalable, dynamic, and remotely accessible grid of processors.

*Note: open to collaborations starting in 2013!