IT 0469 NEURAL NETWORKS

Neuron:A neuron nerve cell is an electricallyexcitable cell that processes and transmits information by electrical and chemical signaling. Chemical signaling occurs via synapses, specialized connections with other cells. Neurons connect to each other to form networks.

Cell Body Contains the nucleus

Dendrites Receptive regions; transmit impulse

to cell body Short, often highly branched May be modified to form receptors

Axons Transmit impulses away from cell

body Axon hillock; trigger zone

Where action potentials first develop

Presynaptic terminals (terminal boutons) Contain neurotransmitter substance

(NT) Release of NT stimulates impulse

in next neuronBundles of axons form nerves

Neurons produce electrical signals called action potentials ( = nerve impulse)

Nerve impulses transfer information from one part of body to anothere.g., receptor to CNS or CNS to effector

Electrical properties result from ionic concentration differences across plasma

membrane permeability of membrane

Single Neuron Physiology

Resting Potential

Inhibitory & Exitatory Action Potential

Nerve cell has an electrical potential, or voltage across its membrane of a –70 mV; (= to 1/20th that of a flashlight battery (1.5 v)

The potential is generated by different concentrations of Na+, K+, Cl, and protein anions (A)

But the ionic differences are the consequence of: Differential permeability of the axon membrane to these ions Operation of a membrane pump called the sodium-

potassium pump

Diffusion of Na+ and K+ down their concentration gradients Na+ diffuses into the cell and K+ diffuses out of the cell

BUT, membrane is 75x’s more permeable to K+ than Na+ Thus, more K+ diffuses out than Na+ diffuses in This increases the number of positive charges on the outside of

the membrane relative to the inside. BUT, the Na+-K+ pump carries 3 Na+ out for every 2 K+ in.

This is strange in that MORE K+ exited the cell than Na+ entered! Pumping more + charges out than in also increases the number

of + changes on the outside of the membrane relative to the inside.

AND presence of anionic proteins (A-) in the cytosol adds to the negativity of the cytosolic side of the membrane

THEREFORE, the inside of the membrane is measured at a -70 mV (1 mv = one-thousandth of a volt)

Number of charged molecules and ions inside and outside cell nearly equal

Concentration of K+ higher inside than outside cell, Na+ higher outside than inside

Potential difference: unequal distribution of charge exists between the immediate inside and immediate outside of the plasma membrane: -70 to -90 mV

The resting membrane potential

Membrane potential is dynamicRises or falls in response to temporary

changes in membrane permeabilityChanges in membrane permeability result

from the opening or closing of membrane channels

Types of channelsPassive or leak channels - always openGated channels - open or close in response to

specific stimuli; 3 major types Ligand-gated channels Voltage-gated channels Mechanically-gated channels

Many more of these for K+ and Cl- than for Na+. So, at rest, more K+ and Cl- are moving than Na+.

How are they moving? Protein repels Cl-, so Cl- moves out. K+ are in higher concentration on inside than

out, they diffuse out.Always open and responsible for permeability

when membrane is at rest.Specific for one type of ion although not

absolute.

Gated ion channels. Gated ion channels open and close because of some sort of stimulus. When they open, they change the permeability of the cell membrane.Ligand-gated: open or

close in response to ligand (a chemical) such as ACh binding to receptor protein. Acetylcholine (ACh) binds to

acetylcholine receptor on a Na+ channel. Channel opens, Na+ enters the cell.

Ligand-gated channels most abun- dant on dendrites and cell body; areas where most synaptic commu-nication occurs

Graded: of varying intensity; NOT all the same intensity Changes in membrane potential that cannot spread far from site of

stimulation Can result in depolarization or hyperpolarization Depolarization

Opening Na+ channels allows more + charges to enter thereby making interior less negative (-70 mV -60mV); see next slide

RMP shifts toward O mV Hyperpolarization

Opening of K+ channels allows more + charges to leave thereby making interior more negative (-70 mV -80 mV); see next slide

RMP shifts away from O mV Repolarization

Process of restoring membrane potential back to normal (RMP) Degree of depolarization decreases with distance from stimulation

site; called decremental spread (see next slide) Graded potentials occur on dendrites and cell bodies of neurons but

also on gland cells, sensory receptors, and muscle cell sarcolemma Affect only a tiny area (maybe only 1 mm in diameter)

If so, how do neurons trigger release of neurotransmitter far from dendrites/cell body?

Voltage-gated Na+ channels sensitive to changes in extracellular Ca2+

concentrations If extracellular Ca2+ concentration

decreases- Na+ gates open and membrane depolarizes.

If extracellular concentration of Ca2+ increases- gates close and membrane repolarizes or becomes hyperpolarized.

Depolarization

Hyperpolarization

Graded potentials decrease in strength as they spread out from the point of origin

Na+ and K+ channels are closed Leakage accounts for small movements of Na+

and K+

Each Na+ channel has two voltage-regulated gates Activation gates – closed in the resting state Inactivation gates – open in the resting state

Some stimulus opens Na+ gates and Na+ influx occurs K+ gates are closed

Na+ influx causes a reversal of RMP Interior of membrane now less negative (from -70 mV -55 mV)

Threshold – a critical level of depolarization (-55 to -50 mV) At threshold, depolarization becomes self-generating

I.e., depolarization of one segment leads to depolarization in the next If threshold is not reached, no action potential develops

Sodium inactivation gates close Membrane permeability to Na+ declines to

resting levels As sodium gates close, voltage-sensitive K+

gates open K+ exits the cell and internal negativity of the

resting neuron is restored

Potassium gates remain open, causing an excessive efflux of K+

This efflux causes hyperpolarization of the membrane (undershoot)

The neuron is insensitive to stimulus and depolarization during this time

1 – RESTING STATE RMP = -70 mV

2 – DEPOLARIZATION Increased Na+ influx MP becomes less negative If threshold is reached,

depolarization continues Peak reached at +30 mV Total amplitude = 100 mV

3 – REPOLARIZATION Decreased Na+ influx Increased K+ efflux MP becomes more negative

4 – HYPERPOLARIZATION Excess K+ efflux

Blue line = membrane potentialYellow line = permeability of membrane to sodiumGreen line = permeability of membrane to potassium

Illustration shows continuous propagation of a nerve impulseon an unmyelinated axon.

• Action potentials occur over the entire surface of the axon membrane.

Most Na+ channels concentrated at nodes. No myelin present.

Leakage of ions from one node to another destabilize the second leading to another action potential in the second node. And so on….

Repolarization Restores the resting electrical conditions of the

neuronDoes not restore the resting ionic conditions

Ionic redistribution back to resting conditions is restored by the sodium-potassium pump

All-or-none principle. No matter how strong the stimulus, as long as it is greater than threshold, then an action potential will occur.

The amplitude of the de-polarization wave will be the same for all action potentials generated.

Sensitivity of area of the membrane to further stimulation decreases for a time

Parts Absolute

Complete insensitivity exists to another stimulus

From beginning of action potential until near end of repolarization.

No matter how large the stimulus, a second action potential cannot be produced.

Has consequences for function of muscle

Relative A stronger-than-threshold

stimulus can initiate another action potential

Faster in myelinated than in non-myelinated In myelinated axons, lipids act as insulation

(the myelin sheath) forcing local currents to jump from node to node

In myelinated neurons, speed is affected by: Thickness of myelin sheath Diameter of axons

Large-diameter conduct more rapidly than small-diameter. Large diameter axons have greater surface area and more voltage-gated Na+ channels

Type A: large-diameter (4-20 µm), heavily myelinated. Conduct at 15-120 m/s (= 300 mph). Motor neurons supplying skeletal muscles and most

sensory neurons carrying info. about position, balance, delicate touch

Type B: medium-diameter (2-4 µm), lightly myelinated. Conduct at 3-15 m/s. Sensory neurons carrying info. about temperature, pain,

general touch, pressure sensations Type C: small-diameter (0.5-2 µm), unmyelinated.

Conduct at 2 m/s or less. Many sensory neurons and most ANS motor neurons to

smooth muscle, cardiac muscle, glands

All action potentials are alike (of the same amplitude) and are independent of stimulus intensity.The amplitude of the action potential is the

same for a weak stimulus as it is for a strong stimulus.

So how does one stimulus feel stronger than another?Strong stimuli generate more action

potentials than weaker stimuli.More action potentials stimulate the release of

more neurotransmitter from the synaptic knob The CNS determines stimulus intensity by

the frequency of impulse transmission

Excitatory signal: Opening of Na+ channelsDepolarizes membrane (-70 mV -60 mV)Brings membrane closer to thresholdMore likely to give rise to an action potential

Inhibitory signalOpening of K+ channelsHyperpolarizes the membrane (-70 mV -80 mV)Takes membrane further from thresholdLess likely to give rise to an action potential

Excitatory postsynaptic potential (EPSP) Depolarization occurs and

response stimulatory Depolarization might reach

threshold producing an action potential and cell response

Inhibitory postsynaptic potential (IPSP) Hyperpolarization and

response inhibitory Decrease action potentials by

moving membrane potential farther from threshold

Individual EPSP has a small effect on membrane potential Produce a depolarization of about

0.5 mV Could never result in an AP

Individual EPSPs can combine through summation Integrates the effects of all the

graded potentials GPs may be EPSPs, IPSPs, or both Two types of summation

Temporal summation Spatial summation

Fig. A illustrates spatial summationFig. B illustrates temporal summationFig. C shows both EPSPs and IPSPs affecting the membrane

Organization of neurons in CNS varies in complexity Convergent pathways: several neurons converge on a single

postsynaptic neuron. E.g., synthesis of data in brain. Divergent pathways: the spread of information from one

neuron to several neurons. E.g., important information can be transmitted to many parts of the brain.

Oscillating circuits: Arranged in circular fashion to allow action potentials to cause a neuron in a farther along circuit to produce an action potential more than once. Can be a single neuron or a group of neurons that are self stimulating. Continue until neurons are fatigued or until inhibited by other neurons. Respiration? Wake/sleep?

Mc-Pitts Model of Neural Networks

How does neuron learn ?

Structure of Processing Element

Signal processing Techniques- Frequency Modulation of signals

Radio Transmission-Frequency domain

Echo suppression in Telephone networks

Frequency response characteristics of different Filters

• Several NN have been proposed & investigated in recent years

• Supervised versus unsupervised• Architectures (feedforward vs. recurrent)• Implementation (software vs. hardware)• Operations (biologically inspired vs. psychologically

inspired)

• In this chapter, we will focus on modeling problems with desired input-output data set, so the resulting networks must have adjustable parameters that are updated by a supervised learning rule

WeightsWeightsWeights

WeightsWeights

WjiVik

F(wji xjF(wji xj

1. Apply input to Adaline input 2. Find the square error of current input Errsq(k) = (d(k) - W x(k))**2 3. Approximate Grad(ErrorSquare) by differentiating Errsq approximating average Errsq by Errsq(k) obtain -2Errsq(k)x(k) Update W: W(new) = W(old) +

2mErrsq(k)X(k) Repeat steps 1 to 4.

Structure of ADALINE

Structure of ALC(Adaptive Linear Combiner)

ALC as a transversal Filter

Use of ADALINE in solving XOR problems

MDALINE Architecture

BPN Architecture

Image to ASCII Conversion using Neural Network

Image to ASCII Conversion using Neural Network (Cont.d)

What is Processing Element. How would you relate the PEs with real neurons

Define Resting Potential. What is the average refractory period of a neuron. Is it limited to a particular value. If Yes mention How?

Differentiate Resting potential and action potential

State Hebbs Learning Rule. Draw a sample memory mapping diagram by your own.

How would you factor out the weight vector from the exception value terms

What is the use of signal processing techniques in neural networks

J. A. Freeman and D. M. Skapura, Neural Networks- Algorithms, Applications and Programming Techniques, Pearson Education( singapore) Pvt. Ltd., 1991.

(Chapters 1 &2) psychology.about.com/od/biopsychology/f/

neuron01.htm www.cell.com/neuron www.neurophys.com faculty.washington.edu/chudler/chnt1.html