Blue Brain

Do neurons dream of electric sheep?

By: Michael R. Rink

Main Points

Objectives of the ProjectData CollectionModelExpected ConsequencesFuture Work

Objectives

Phase 1: 1cm^3 NCCJune 6, 2005 – November 26, 2007Detailed Cellular scale (10,000 neurons)

Full NeocortexModularAtomic (days/ms), Genetic (another 5-10

years)Cellular – 100,000 Point Source -100 million (mouse brain)

Full Mammal Brain 2 years Emergent behavior

Learning/ self-wiring

Interact with environmentJapanese built robot ratWhole new models to connect sensory cortex

to hardware

Full Human Brain Need speed increase of ~10million fold Algorithm improvements 100-1,000 fold ASICS 1,000-10,000 fold Multi-core & Multi-threading are really

useful

On Going Data Collection

Decades of research on 2 year old rat somatosensory cortex

Hundreds of new data points a day

50% success rate Fig A. Patch Clamp Robot

Blue Column 1The microcircuits (from left toright) for layers 2, 3, 4 and 5

Blue Column 2

A single thick tufted layer 5 pyramidal neuron located within the column.

c | One pyramidal neuron in layer 2, a small pyramidal neuron in layer 5 and the large thick tufted pyramidal neuron in layer 5.

d | An image of the NCC, with neurons located in layers 2 to 5.

Blue Column Construction

Model the 3d neuron Repair defects from in vitro preparation

Patch clamping damages the cell Insert 12+ ion channels in each

component compartment based on HH models & studiesOctober 2005 Petilla Classification

Convention

Each neuron is unique within statistically determined bounds – procedurally generated

Stored in database after creation Each layer procedurally generated from

database then shifted/rotated to avoid collisions Generating the column takes longer than running it Simulation runs at 1/100th real time

Individual synapses functionalized and plasticized with learning algorithms

Map Neuron to Processor 1:1 – 10:1 Axons mapped onto inter-processor communications

Expected Consequences

Each NCC will be uniqueHopefully, NCC output will approach

convergenceNCCs will begin to appear similar (lack of

molecular detail to tie in genetic elements) Connected Neurons fire together

Predicted by Hebbian learningEmergent behavior

Blue = rest Yellow = group 1 Red = group 2

Global Excitatory state

Notice the blue, inhibited neurons

Future Work

Experimental Aid How complex of a problem can an NCC

solve Emergence of Conciousness

Director Markram

Any Questions?