+ All Categories
Home > Technology > Artificial brains

Artificial brains

Date post: 31-Oct-2014
Category:
Upload: serendipity-seraph
View: 726 times
Download: 6 times
Share this document with a friend
Description:
 
11
Artificial Brains
Transcript
Page 1: Artificial brains

Artificial Brains

Page 2: Artificial brains
Page 3: Artificial brains

Memristors

• More unified memory and processing per element– Synapse like

• Lower power and high parallelism for some brain-like models

• DARPA is funding a lot of this research– Active goal of powerful, cheap, low power

adaptive intelligence– Probably deployed on battlefield

Page 4: Artificial brains
Page 5: Artificial brains

Memristor Characteristics

• Simpler than transistor• Smaller (3 nm vs 30 nm)• Stateful logic– Computation + memory characteristics– Remembers last voltage

• Similar to brain neuron / synapse

• Non-volatile – Ultra low power

• Sub-nanosecond switching speed

Page 6: Artificial brains

Why are animal brains so efficient?

• Human brain runs @ 20 – 30 watts• Blue brain scalable would require own GW power

plant• Difference may be fundamental architectural units.• It the mammalian brain storage and processing

happen in the same place. • Brain circuits can operated at 100 millivolts. Most

CMOS at around 1 volt.• Memristors rival this power efficient, density, and

unity of processing and storage.

Page 7: Artificial brains

Blue Brain

• Goal is emulated functioning slice of rat brain• Development of generic facility to rapidly model

and simulate any brain region• Model of neocortical column containing 200 types

of neurons• Attempt to reverse engineer brain – But not to create a brain (!?)– Nope. New goal to build functioning, artificial human

brain in next 10 years. Markham claims this is possible.

Page 8: Artificial brains

IBM’s “cat brain”

• Cell-by-cell simulation of human visual cortex– 1.6 billion virtual neurons, 9 trillion synapses

• Size of cat’s brain

• May be key to interpreting biological sense information (sight, hearing, touch)

• Seems to have lacked neural patterning– Apparently takes wandering environment

• Perhaps embodiment. Experimental data supports this.

• Eventual goal – simulating entire human cortex• The approach badly needs memristors or better

– Very very power hungry and lots of heat

Page 9: Artificial brains

DARPA Synapse

• Systems of Neuromorphic Adaptive Plastic Scalable Electronics

• Eventually want full human equivalent artificial brains, starting with cat like intelligence

• Many functions of cat brains are very important for robotics bio-equivalent intelligence

• Brain body size ratio mysetry

Page 10: Artificial brains

De Garis Brain

• Evolving neural network modules for specific task clusters

• Integration of these is the “art” part• 3D cellular automata based– Memristors should help– Possible to run on modern GPUs?

• “China Brain Project”– 4 year project linking 10,000 – 50,000 NN modules– Brain with thousands of specialized pattern detectors

Page 11: Artificial brains

Numenta – Jef Hawkins

• Mathematical model of what he believes is fundamental brain algorithm

• Hierchical temporal memory (HTM) – Biomemetic model based on memory-prediction

chains– Combines and extends parts of bayesian networks,

spatial and temporal clustering algorithms and uses a NN like hierarchy of nodes


Recommended