IBM Systems Cognitive Systems Dr. Wolfgang Maier Director HW - - PowerPoint PPT Presentation

• Dr. Wolfgang Maier

Director HW Development IBM Research & Development wmaier@de.ibm.com 9/2017

IBM Systems Cognitive Systems

IBM Research & Development

Forschung Hardware-Entwicklung Software-Entwicklung Hardware- und Software-Entwicklung Austin Böblingen Zürich Tucson San Jose Hursley Greenock Rom Endicott East Fishkill Burlington Poughkeepsie Watson Toronto Rochester Boulder Raleigh Krakau Paris Fujisawa Yasu Indien Tokio Yamato Haifa Perth Bangalore Gold Coast Sydney Pune Shanghai Taipei China Beaverton São Paolo Kairo Moskau Minsk Costa Mesa Foster City Vancouver Peking Dublin Santa Teresa Almaden La Gaude

Digitale Transformation

Powered by data

Photo - Lithography UV – Lithography 193 nm eUV – Lithography 13 nm

i

Output Input

c

Artificial Intelligence

Artificial Intelligence

Training Architecture supervised unsupervised neuromorphic von Neumann

True North Zeroth Spinnaker CAL HTM CNN

?

Neural Networks

Stochastic Gradient Descent

Quantum Computing

Spiking Neurons

Prominent Features of Spiking Neurons after Izhikevich

Neuron Function

• Emulation of analog behaviour by +/- 255 INT variable
• 2-dimensional on-chip synaptic weighted network and off-chip packet

based thru-neuron routing for multi-chip scaling

• Update of Synaptic network every ms (logical / biological clock), internal

processing ~ 1MHz

• Neuron fires a spike (45 pJ) to the network if in the last update cycle a

threshold was reached or exceeded

• Stochastic and leak behaviour configurable

Vj(t) = Vj(t-1) + * [(1 - cj) *kj + sign(kj) * cj * F(|kj|, qj)]

+1

• 1

( ) Vj(t) = Vj(t-1) + S Ai(t) * zij * [(1 - bj) *sj + sign(sj) * bj * F(|sj|, qj)] + Leak

i=0 255

Membrane potential for neuron j at time t Leak

OUTPUTS:

Predictions Context Stable Concepts (SDR) Motor commands

INPUT:

Spatial-temporal data streams of any kind

“ Universal Cortical Engine “

Sparse Distributed Representations (SDR)

17

Find semantic similarities of words in Wikipedia

Document corpus

(e.g. Wikipedia)

128 x 128

100K “Word SDRs”

minus minus minus minus = Apple Fruit Computer

Macintosh Microsoft Mac Linux Operating system ….

runners up were

Big Data Content Analytics

IBM Technology Depth

Business Analytics Databases / Data Warehouses 2880 Processing Cores 16 Terabytes Memory (RAM) – 20TB Disk

System Specifications

90 IBM P750 Servers 80 Teraflops (80 trillion

• perations per second)

Workload Optimized Systems

IBM Watson

Synapse Hardware

TrueNorth Technology 28nm Year 2012 Transistor Count 5.4 billion Power 0.05W

Intelligence

Errorfunctions

T es t-E rro r d u rin g tra in in g C ro s s

• E

n tro p y E rro r (C ) d u rin g tra in in g

Quantum Computing