Transcript MIT lecture
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 technique blood 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 lightconduction 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 nondigital 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 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. 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