4-Zoology-Organ-Nerve-Brain-Computation

brain computation

Brain performs computations {brain computation}. Brain scanning and brain lesion analysis shows that brain regions mediate brain functions {localism}. Brain regions have functions, and brain functions involve several brain regions coordinating sequentially and simultaneously [Carter, 1999].

input and output

Neuron populations receive input-signal sets and calculate output-signal sets. For functions, brain has several paths that make same output from same input.

All central-nervous-system input goes to cerebellum, reticular formation, and thalamus. Input that can eventually cause sense qualities goes to sense neurons in thalamus and then cerebrum. Brain inputs and outputs coordinate. For example, motor neurons and muscle receptors link.

input and output: global information

All brain information uses many neurons, which process only local information, but result is global [Black, 1991].

input and output: convergence

Inputs from different cortical regions {convergence region, vision} converge on cortical regions {convergence zone}. Convergence-zone output includes feedback to input cortical regions. Brain has more than 1000 convergence regions and more than 100 convergence zones.

input and output: distributed processing

Brain has many recurrent, lateral intracortical connections, and neurons receive from large spatial area. No neuron has simple direct circuit [Abbott et al., 1996] [Dayan and Abbott, 2001] [Rao et al., 2002] [Rolls and Deco, 2002] [Shadlen et al., 1996] [Zhang et al., 1998].

More than 50% of cortical neurons send output to distant cortical areas.

input and output: important information

Brain gives command to region with the most-important information.

coding

Information theory applies to neural coding, and one impulse can carry more than one information bit [Rieke et al., 1996] [Shannon and Weaver, 1949].

coding: repetition

Brain cells typically experience same correlations, associations, and comparisons many times. Processing in small brain region repeats often. Repetition improves efficiency as cells modify.

coding: electrochemical coding

Axon ion flows change voltage, and receptor-membrane neurotransmitters change voltage, so neuron coding is electrochemical.

coding: discrete coding

Neurons generate neurotransmitter packets and action potentials, rather than graded potentials, so information flows are discrete on-or-off sequences rather than continuously-varying voltages or currents.

coding: frequency code

Neuron impulse frequency indicates both time interval and intensity. Spike frequency F above normal frequency NF is constant C times power p of receptor displacement X from normal N: (F - NF) = C * (X - N)^p.

Neuron spike frequency never exceeds 800 spikes per second and never becomes 0. At rest, frequencies can be few per second or hundreds per second.

coding: information amount

If axon-hillock depolarization or no depolarization can happen every millisecond, millisecond intervals are either on or off. In 100 milliseconds, 2^100 bits can travel down axon. 2^100 bits is approximately 10^30 bits, which is approximately 1000^10 bits.

If number of possible letters, digits, and punctuation marks is 1000, 1000^10 bits can represent ten-letter words, labels, indexes, or pointers, so one neuron acting for 100 milliseconds can code for any ten-letter word. One hundred thousand neurons acting for 100 milliseconds can code for any million-letter string. One hundred thousand neurons acting for 100 milliseconds can code for any million-dot image, if dots can have 1000 levels of gray and color.

One hundred neurons acting for one millisecond can code for any ten-letter word. One neuron acting for 10^7 milliseconds, approximately 3 hours, can code for any million-letter string or million-dot image.

coding: summation

Axons depolarize for percentage of time. In time intervals, such as ten milliseconds, so short that receptors do not have time to change, depolarization sequences with same total number of depolarizations have equivalent physiological effects, no matter in what order signals arrived. In these cases, axon coding is not a significant factor, and irregular conduction rates do not matter. Ten-millisecond intervals have 10 instants that can be ON or OFF, and so intervals have 2^10 different possible sequences. However, such short time intervals have only 11 physiological possibilities: 0 to 10 total depolarizations.

computations

Neurons promote, amplify, block, inhibit, or attenuate signals. Neuron circuits can lower thresholds, switch signals, amplify signals, filter frequencies, set thresholds, control currents, direct flows, induce currents, control voltages, compare signals in time and space, add quantities, multiply quantities, and perform logical operations.

Most visual cortical neurons respond to line contrasting with background. Output maximizes if contrast is in receptive-field middle. Response falls off rapidly as contrast line moves to either side. Function looks like Gaussian distribution with large variance or looks like two sigmoidal functions joined at maximum. Most neurons behave like Gaussian HBF units, rather than multilayer-perceptron (MLP) sigmoidal units.

computations: potentiation

Nitric oxide from receptor neuron maintains sending-neuron long-term potentiation (LTP). Cyclic AMP second messenger sensitizes receptor neuron. NMDA glutamate receptors sensitize receptor neuron. Potentiation makes neuron reach threshold with lower input than normal excitation level.

computations: association by neurons

Interneuron can detect stimulus timing and amplitude between two pathways. Two pathways can both contact an interneuron that controls secondary circuit. If both pathways carry current simultaneously, secondary circuit maintains current. If both pathways carry no current or only one has current, secondary circuit has no current. Interneuron detects simultaneous pathway activation.

computations: conditional statement

Depolarization can happen only if all inputs to neuron are active. Depolarization can happen if one input to neuron is active. Depolarization can happen only if some inputs are active and some inactive.

computations: difference

Neurons compare signals to detect change or difference, rather than absolute values.

computations: feedback

Neuron circuits with excitatory and inhibitory feedback have short, synchronous discharges in local clusters.

computations: indexing system

Brains have index list to rapidly locate information.

computations: inhibition

Inhibition dampens neuron signals and more quickly returns neurons to resting states, allowing shorter time intervals and quicker responses. Between two stimuli, inhibition increases stronger signal and decreases competing weaker stimulus. Thus, inhibition can enhance weak stimulus that has only weaker stimuli nearby in space or time.

computations: inverse model

If supplied with sample responses, inverse model can predict input commands.

computations: movement initiation

Movement initiation and programming are separate asynchronous processes. Deciding to perform action corresponds to building positive feedback in limb premotor network.

computations: movement programming

Movement initiation and programming are separate asynchronous processes. Cerebellar cortex programs which movement to perform. Programming can happen while preparing movement or moving.

computations: positive whole numbers

Synaptic transmission uses neurotransmitter-molecule packets. Axon depolarizations are independent and countable. Therefore, neurons use whole number calculations and whole number ratios, with no fractions or real numbers. Because neurotransmitter molecules and axon depolarizations have no opposite, neurons use only positive numbers, never negative numbers.

computations: synchronizing

Neuron reciprocal interactions can synchronize signal phase, if reciprocal signal delay is less than one-quarter oscillation. Neuron excitatory reciprocal connections can find feature relations by enhancing, prolonging, and synchronizing neuron responses.

computations: timing

Brain uses time-delay circuits, inhibitory signals, and circuit-path rearrangements to time neuron events.

computations: transformations

Brain can use geometric ratios, translations, rotations, and orientations to represent distances and manipulate lines and shapes.

validation

Brain sends inputs along several paths to cross-validate processing and representations by comparing redundant process outputs.

association triad neuron mechanism

Interneurons can associate reflex-pathway nerve-cell states with other reflex-pathway nerve cell states {association triad neuron mechanism}.

interneurons

Nerve cells in pathways from sensation to action connect using interneurons for cellular associative learning. Interneurons laterally excite or inhibit main neurons.

process

Cell that is to learn inhibits interneuron. Interneuron excites learning cell and inhibits paired cell. Paired cell inhibits interneuron and learning cell.

Associative learning requires asymmetric connections. Both main neurons inhibit interneuron, but interneuron excites learning neuron and inhibits paired neuron. Paired cell inhibits learning cell, but learning cell does not synapse on paired cell.

process: input

Association requires simultaneous stimuli to both sense cells.

process: circuit

After stimuli cease, asymmetric association-triad circuit causes signals to continue to flow around neurons, keeping learning-cell electric potential smaller. This makes it easier to stimulate learning cell.

When paired stimuli end, paired cell stops inhibiting interneuron and learning cell, so interneuron excites learning cell more. Learning cell has decreased inhibition and so inhibits paired cell even more, keeping state going until unpaired stimuli disrupt positive feedback, so neurons return to normal.

simplicity

Association nerve triad is simplest associative learning system. One neuron alone can only sensitize or desensitize itself. Two neurons can only have reverberation. No other three-neuron arrangement can make a learning circuit.

Interneuron must have inhibition by path neurons, so interneuron voltage increases when stimulation stops. Interneuron must stimulate learning cell, so learning-cell voltage increases after stimulation stops. Paired cell must inhibit learning cell, so learning-cell voltage increases after stimulation stops. Interneuron must inhibit paired cell, so, as interneuron increases voltage, paired-cell inhibition decreases.

There must be no connection from learning cell to paired cell. If there is inhibition, stimulation end increases paired-cell inhibition on learning cell. If there is excitation, stimulation end decreases inhibition, but excitation mimics stimulation, so there can be no excitation between direct paths.

There is no interneuron excitation, because stimulation end only quiets them.

uses

Association does not detect current or no current in one or the other pathway, only simultaneous inputs.

uses: negative association

If sense cell inhibits learning cell, circuit still works by positive feedback but in reverse, so voltage difference becomes more. This makes learning cell more difficult to stimulate.

uses: attention

Association allows attention-like input acknowledgement.

uses: reflex control

Reflexes receive signals from sensors and activate muscles. Association triads can control reflexes by signals from other nerve pathways or from brain.

uses: time

Association triads can detect simultaneity in time.

uses: space

Association triads can detect simultaneity in space. Time can code spatial distance.

uses: intensity

Association triads can compare intensities. Time can code intensity.

labeled-line code

Nerve pathway {labeled-line code} itself indicates stimulus nature and spatial location.

neuromodulatory system

Cortex mechanisms enhance and suppress synapses and redistribute axon terminals among receptors {neuromodulatory system}. First, synapses receive non-specific input from cholinergic and noradrenergic neurons. Later, they receive input from both body sides [Crick and Koch, 1998].

redundancy of potential command

Command passes to region with the most-important information {principle of redundancy of potential command} {potential command redundancy} {redundancy of potential command}.

re-entry

Cortical-neuron axons from sense brain areas can re-enter brain areas {re-entry} {re-entrant pathway}. Re-entry synchronizes and coordinates but does not provide feedback. Most nerve pathways send signals back to starting points after one or more synapses. Memory depends on brain pathways that re-excite themselves.

Smith predictor

Control system models {Smith predictor} can predict responses if supplied with sample commands. Smith predictors combine delayed and undelayed control system models to build controller for systems with large time delays.

4-Zoology-Organ-Nerve-Brain-Computation-Regions

lateral inhibition

Nearby neurons inhibit neuron {lateral inhibition}|, to increase contrast.

spreading activation

Excitation tends to spread through region {spreading activation} {spreading excitation}, to link regions.

Related Topics in Table of Contents

4-Zoology-Organ-Nerve-Brain

Drawings

Drawings

Contents and Indexes of Topics, Names, and Works

Outline of Knowledge Database Home Page

Contents

Glossary

Topic Index

Name Index

Works Index

Searching

Search Form

Database Information, Disclaimer, Privacy Statement, and Rights

Description of Outline of Knowledge Database

Notation

Disclaimer

Copyright Not Claimed

Privacy Statement

References and Bibliography

Consciousness Bibliography

Technical Information

Date Modified: 2022.0225