230 likes | 363 Vues
This presentation explores the human brain as a sophisticated computing system, highlighting its unique capabilities and how they differ from digital computers. It covers essential aspects of brain structure, neuron connections, and processing speed, showcasing the brain's parallel architecture and vast storage potential. We also discuss the inherent trade-offs in human cognition and compare biological computing to digital systems, emphasizing the lessons we can learn from the brain's evolution and functionality to inspire future bio-inspired computing technologies.
E N D
Computing Architectures The human brain as computing system Based on presentation from http://www.stanford.edu/class/symbsys100/ and http://www.willamette.edu/~gorr/classes/cs449/brain.html
Plan • From symbols to meat • Meet the brain • Brains vs. digital computers • Bio-inspired computers • Reasoning module: concluding discussion
Motto • Human cognition is based on a very specific computing system, with specific limits, inherent trade-offs, etc. that are not necessarily the same as for digital computers • It is therefore worth looking at the "mind's implementation" in order to learn more about the limits of our mind/cognition
Plan • From symbols to meat • Meet the brain • Brains vs. digital computers • Bio-inspired computers • Reasoning module: concluding discussion
The brain – just 2 pounds of meat? • The cortex • 1.3-1.4kg (2% of the body weight) … [13,14] • 2,500 cm2 (rat: 6 cm2, elephant: 6,300 cm2)[14] • 1,300-1,500 cm3 • 2 hemispheres connected by corpus callossum (250 mill. nerve fibers) • Inputs: • spinal cord • optic nerve (1.2 mill.) • cranial nerves (12) • auditory system, …
The lobes • 4 lobes: occipital, parietal, temporal, frontal • Occipital: vision • Parietal: touch, pressure, temperature, pain • Temporal: auditory information, long term memory • Frontal: short term memory, planning, emotion, movement… • Biggest difference from our closest evolutionary ancestors
Taken from http://www.sciencebob.com/lab/bodyzone/brain.html
Neurons [14] • 100 billion neurons (children) • 300 million – octapus; • 18,000 – sea slug Aplysia; 350 - leech • Diameter: 4 – 100 microns • Weight: 10-6 grams • Length: <1 mm – 4 feet (in the leg) [15] • Length of Giraffe primary afferent axon: 15 feet • Loss of neurons: ~1/sec 31 million/year an octapus/10 years • ~5,400 at the end of this lecture (sorry!)
How do we know? • Non-invasive (1mm3 ~ 6-7*104 neurons) • EEG (Electroencephalogram), • ERP (Early receptor potential) • fMRI (blood flow; ~1mm; secs-mins) • MEG (Magnetoencephalogram with ERP: ~1.5mm; msecs-secs) • PET (imaging techniqueblood flow; 1mm; >mins) • Invasive methods: electrodes (1 neuron; msecs) • Lesions • Permanent: injury, disease • Temporary: specific drugs, TMS (<1mm; <secs) • All methods have trade-offs (spatial, temporal resolution)
The brain as a computational system • The brain is • biological • de-central (plasticity) • non-digital • highly parallel • What does this mean?
The brain: a biological CS • not manufactured from scratch with a certain intention in mind, but subject to evolution • Co-adaptation; its parts must have been of use • Not made out of copper or light-conduction cables .... slow • Signal speed: MAX=120m/s, AV.=6.5m/sec (1.2 - 250mph) [14] • Signal frequency: up to 1000Hz (activ./sec)
Non-digital • At least to some degree, the brain is non-digital • On the lowest level (i.e. within the neuron): quasi-digital • this creates an analog signal travelling along the neuron • at the synapses this is converted into a chemical signal, which in turn triggers an elecrical signal.
The brain: a highly parallel CS • Some neurons have up to 150,000 connections (others as low as 2) • average: 1,000-10,000 [14] • different brain regions are highly interconnected • human can manage many tasks at the same time (sitting, listening to the lecture, doodling) • however, there are also parts of the brain which are involved in a lot of tasks "narrow passages" for computation
Plan • From symbols to meat • Meet the brain • Brains vs. digital computers • Bio-inspired computers • Reasoning module: concluding discussion
Storage capacity of the brain - I • 100 billion neurons • 1011! hypothetically possible connections • upto 150,000 connections between each neuron (180,000km of myelinated nerves) • during the first year of life, the child generates ~ 15,000 connections for each neuron (during growth: 250,000 per minute!) • “… this program will support more than 130,000 [i.e. 1.3 * 105] neural connections…”
Storage capacity of the brain - II • # bits = # of neurons * # of connections • 1 * 1011 * 1.5 * 105 = 1.5 * 1016 bits • The entire Enc. Britannica contains 109 bits of information (Turing 1950) • In 1987, Hideaki Tomoyori memorized the first 40,000 digits of π
Information processing speed of the Brain • # bits/sec = # ops/sec* # bits/op • 10 ops/sec per synapse (connection) [3,4] • ~1.5 * 1017 bits/sec information transfer • Estimates of the brain's computing power range from 1011 to 1020 bits/sec • Converging evidence for ~ 1015 [2,3,5,9,14] • ~100 teraflops (8 bit words); ~ 8 teraflops (128 bits words)
Brain vs. digital Computers • Fastest computer atm: • 40 terra flops(5,000 processors; NEC) • Planned • 360 terra flops (130,000 processors; IBM) • ~ 3-4 times faster than the human brain (8 bit words); 40 times faster otherwise.
Plan • From symbols to meat • Meet the brain • Brains vs. digital computers • Bio-inspired computers • Reasoning module: concluding discussion
Bio-inspired models of computation • This gives us a motivation to investigate bio-inspired models of computation • Learn about the brain by modeling it • Take advantage of billions of years of evolutionary design • Develop robust computational systems • Neural networks
So what? • “It is true that a discrete-state machine must be different from a continuous machine. But if we adhere to the conditions of the imitation game, the interrogator will not be able to take any advantage of this difference.” Turing (1950:451)
References • Gazzaniga, Ivry & Mangun (1998): Cognitive Neuroscience. The Biology of the Mind. Norton. • Merkle, Ralph C. (1988): How many bytes in human memory? at http://www.merkle.com/humanMemory.html • Merkle, Ralph C. (1989): Energy Limits to the Computational Power of the Human Brain; at http://www.merkle.com/brainLimits.html • Principles of Neural Science, by Eric R. Kandel and James H. Schwartz, 2nd edition, Elsevier, 1985 • http://www.coping.org/earlyin/ruleout/reason.htm • http://www.jsmf.org/zarticles&pap/John/neural_connections.htm • http://ifcsun1.ifisiol.unam.mx/Brain/neuron.htm • http://ifcsun1.ifisiol.unam.mx/Brain/neuron2.htm • http://www.rfreitas.com/Nano/DeusExDigita.htm • http://www.cheshireeng.com/Neuralyst/nrlnds.htm • http://www.top500.org/ • http://www.consciousness.arizona.edu/hameroff/ • http://www.neurologicalalliance.org.uk/pages/network/answers.asp • http://faculty.washington.edu/chudler/facts.html • http://www.uncc.edu/sspauldi/LECNote/ch02.html