2MARK

16
!"" Two mark question and answers 1. What is a neural network? Neural network is a massively parallel distributed processing system, made of highly inter-connected neural computing elements that have the ability to learn and thereby acquire knowledge and make it available for use. 2. What are the advantages of neural networks? Non-linearity makes them usable for non-linear applications They have input-output mapping capability Adaptivity Fault-tolerance VLSI implementability Uniformity of analysis and design Neuro-biological analogy Robust Flexible Ability to deal with variety of data situations Collective computation 3. Sketch a biological neural network? 4. Compare biological and computer (artificial) neural networks? Artificial neural network Biological neural network Fast in processing slow in processing Simple in size complex in size New information replaces old data no replacement of old data Sequential operation parallel operation Fault tolerance is not satisfactory fault tolerance is very good

description

soft computing

Transcript of 2MARK

  • !""

    Two mark question and answers

    1. What is a neural network?

    Neural network is a massively parallel distributed processing system, made of highly

    inter-connected neural computing elements that have the ability to learn and thereby

    acquire knowledge and make it available for use.

    2. What are the advantages of neural networks?

    Non-linearity makes them usable for non-linear applications

    They have input-output mapping capability

    Adaptivity

    Fault-tolerance

    VLSI implementability

    Uniformity of analysis and design

    Neuro-biological analogy

    Robust

    Flexible

    Ability to deal with variety of data situations

    Collective computation

    3. Sketch a biological neural network?

    4. Compare biological and computer (artificial) neural networks?

    Artificial neural network Biological neural network

    Fast in processing slow in processing

    Simple in size complex in size

    New information replaces old data no replacement of old data

    Sequential operation parallel operation

    Fault tolerance is not satisfactory fault tolerance is very good

  • !""

    5. Mention some applications of neural networks.

    Signal processing (noise suppression, echo cancellation)

    Control applications

    Pattern recognition and classification

    Medical applications

    Speech re-production

    Speech recognition

    6. What is an activation (squashing) function?

    It is a function used to limit (squash) the amplitude of the output of a neuron.

    7. Mention the activation functions used by the neural networks.

    The activation functions usually applied for artificial neural networks are:

    Threshold function (binary step function)

    Binary sigmoid function (logistic sigmoid)

    Bipolar sigmoid function (hyperbolic tangent)

    Piece-wise linear function

    Identity function

    8. Define identity activation function.

    9. Define binary step activation function.

    The binary step activation function, commonly referred to as threshold function or

    Heaviside function is defined as

    The threshold value can be a non-zero value (), in which case, it is defined as

    10. Define binary sigmoid activation function.

    The binary sigmoid / logistic function is defined as

    , is the steepness parameter

    #

    #

    $%#&

  • !""

    11. What is learning?

    Learning is a process by which the free parameters of a neural network are adapted

    through a process of stimulation by the environment in which the network is embedded.

    12. What are the various learning methods used by artificial neural networks?

    Learning methods used by ANN are:

    Supervised learning (learning with a teacher)

    Unsupervised learning (learning without a teacher)

    Re-inforcement learning

    13. What is supervised learning?

    In this type of learning, the teacher provides the net with the desired response. When

    input is applied to the net, it produces an output. This output is compared with the desired

    response given by teacher, and the error is calculated. Based on this error, the net is

    trained.

    14. What is unsupervised learning?

    In unsupervised learning, there is no teacher to present the desired response. Hence the

    neural network has to learn on its own. The network learns of its own by discovering and

    adapting to structural features in input patterns.

    15. What is reinforced learning?

    This is a type of learning where the teacher just says whether the output is correct or not.

    Based on this, the net continuously interacts with the environment and corrects itself

    based on some index of performance.

    16. What are the learning rules used by ANNs?

    The commonly used learning rules are:

    Hebbian learning rule

    Perceptron learning rule

    Delta learning rule

    Widrow-Hoff learning rule

    Correlation learning rule

    Winner-take-all learning rule

    Outstar learning rule

    17. What are the requirements to be satisfied by learning rules?

    The learning should lead to convergence of weights.

    The learning time should be as small as possible.

    An on-line training is preferred to off-line training. I.e. the weights should be

    adjusted on presentation of each sample and not separately.

    Learning should use only local information as far as possible. I.e. the change in

    weight on a connecting link between two units should depend on the states of the

  • !""

    two units only. In such a case, it is possible to implement the learning law in

    parallel for all weights, thus speeding up the learning process.

    18. What is Hebb rule?

    According to Hebb rule, learning occurs by modification of the synaptic strengths

    (weights) in a manner such that, if two interconnected neurons are both on or off at

    the same time, then the weight between those neurons should be increased.

    19. Explain the perceptron learning rule.

    The weight change depends on the desired output. The rule is expressed as:

    ! " # " Where is the learning rate ( $% ) t is the target output or desired output

    20. Explain delta learning rule.

    The delta rule changes the weights of the neural connections (synaptic weights) so as to

    minimize the difference between the net input to the output unit (y_in) and the target

    value t. The aim is to minimize the error. The rule is expressed as:

    ! " & ' and ( ( ! " & '

    21. Give the expression for Widrow-Hoff learning rule.

    " & ) 22. Explain correlation learning rule.

    The rule states that ift is the desired response due to an input xi, then the

    corresponding weight increase is proportional to their product. The rule can be

    expressed as:

    " 23. Where do winner-take-all neural nets find application?

    They are used to learn statistical properties of inputs.

    24. Give examples of some single-layer nets.

    Some examples of single-layer feed-forward neural networks are:

    Mc-Culloch Pitts neuron

    Hebb net

    Perceptron

    Adaline

  • !""

    25. Draw the architecture of Mcculloch-Pitts model of neural network.

    26. What is a linear separable problem?

    If there are weights so that all of the training input vectors for which the correct response

    is +1 lie on one side of the decision boundary and all the training input vectors for which

    the correct response is -1 lie on the other side of the decision boundary, the problem is

    said to be linearly separable.

    27. What is the drawback of Hebb net?

    Hebb net Cannot be applied for binary data. (This method cant distinguish between a

    training pair in which input is on, target is off and a training pair in which both input

    and target are off )

    28. What are the advantages of Perceptron?

    Weights converge to correct values.

    Learning decreases as more training patterns are presented to the network.

    29. Mention the applications of Perceptron.

    Perceptrons find application in:

    o Implementation of logic functions o Character recognition o Pattern classification o Classification of noisy patterns..etc.

    30. What is ADALINE? An ADALINE (ADAptive LInear NEuron) is a single unit (neuron) that receives input

    from several units. It also receives input from a unit whose signal is always 1.

    #

    #

    #'

    ()

    #'*

    ()

  • !""

    31. What is a single layer net? A single layer net is one which has only one layer of computational neurons. The

    computational node is usually the output node in case of single layer net.

    32. What is a multi-layer net? A multi-layer network is one which consists of one or more layers of computational

    neurons. Usually the computational capability is processed by hidden layer neurons. So in

    multi-layer nets, in addition to output node, hidden layer of neurons is used to improve

    the computational capability of net.

    33. What is the need of hidden layers? Hidden layers are computational layers, in the sense that, some sort of computations are

    performed in these layers. Hidden layers are usually used to improve the computational

    capability of the network.

    34. What is MADALINE? MADALINE is Multi-ADALINE, in that it is the multi-layer version of ADALINE. This

    has a good computational capability compared to the ADALINE.

    35. What is Back-propagation algorithm? Back-propagation algorithm is a powerful algorithm for training Multi-Layer Perceptrons

    (MLPs). Here, the error is fed back to the layers preceding it.

    36. What are the stages involved in training of networks by back-propagation? The training of network by back-propagation involves three stages. They are:

    Feed-forward of the input training pattern,

    Calculation and back-propagation of associated error, and

    Adjustment of weights.

    37. Why are multi-layer nets preferred over single-layer nets?

    Multi-layer nets have the capability of learning any complex input-output mapping,

    whereas single-layer nets are limited to simple input-output mapping.

    38. Give the architecture of back-propagation net.

    +

    ,

    -

    + +*

    ,. ,)

    - -

    /

    0

    00.

    0)

    /)*

    0)

    /. //*

    11111111111111111111

  • !""

    39. What are the characteristics expected out of activation function for back-propagation nets?

    An activation function for the back-propagation net should have the following

    characteristics:

    It should be continuous

    It should be differentiable

    It should be monotonically non-decreasing

    Its derivative should be easy to compute

    40. What are the activation functions used by back-propagation nets?

    Usually, bipolar sigmoid and binary sigmoid are used as activation functions.

    41. What are the factors affecting the convergence of training of neural networks?

    The important factors affecting the convergence of training are:

    A. Initial weights

    B. Weight adjustment mechanism

    C. Activation function

    D. Selection of learning constant

    E. Momentum method

    1. What is a feedback neural network?

    Feedback neural nets, also referred to as recurrent networks, are networks with at least

    one feedback loop.

    2. What is the use of unit delay elements in feedback networks?

    Unit delay elements (z-1

    ) are used to introduce non-linear dynamical behavior into the

    network.

    3. What is an associative memory neural network?

    Associative memory neural nets are single-layer nets in which the weights are determined

    in such a fashion that the net can store a set of pattern associations, each association

    being an input-output pair s: t, and is able to recall the desired response pattern when

    given an input stimulus that is similar, but not necessarily identical to the training input.

    4. What is the difference between auto-associative and hetero-associative memory

    neural networks?

    In auto-associative memory net, each output vector t is same as the input vector s with

    which it is associated, whereas in hetero-associative nets, thets are different from

    thess.

  • !""

    5. What is a Hopfield neural network?

    Hopfield nets are iterative, fully connected, auto-associative nets having symmetrical

    weights with no self connections, used for pattern association applications.

    6. What is Energy function or Lyapunov function?

    It is a function used to prove the convergence of the Hopfield

    net.* &+,- - ./. .. & - ! - 0 is the commonly used Lyapunov function.

    7. Draw the Hopfield net.

    8. Give an expression for storage capacity of Discrete Hopfield network.

    According to Hopfield, the number of binary patterns that can be stored and recalled with

    reasonable accuracy is1 2 +,, where n is the number of neurons in the net. The number of bipolar patterns that can be stored and recalled is 1 34 5678 3

    9. What is a Content Addressable Memory (CAM)? Content Addressable Memory are those that are able to retrieve a pattern (item) stored in

    memory in response to the presentation of an incomplete or noisy version of that pattern

    (item).

    10. What is vector field method used for? Vector field method is a numerical method for determining the transient response of

    continuous time networks.

    11. Mention some applications of neural networks.

    Artificial neural networks are used in almost all areas. Some applications include:

    Signal processing Control Pattern recognition Medicine Speech production and recognition Business

    + + + +

  • !""

    12. Draw a schematic for forward plant identification.

    13. Mention the advantage and drawback of forward plant identification. The advantage of forward plant identification is that it is feasible. However, the drawback

    is that this method does not allow for immediate construction of the plant controller.

    14. Draw a schematic of plant inverse identification.

    15. Mention the advantage and drawback of plant inverse identification. One advantage of this method is that it facilitates simple plant control. However it suffers

    from the drawback that the identification is not always feasible. Also problem arises in

    this method when the plant inverse is not uniquely defined.

    23

    3 "4

    "5

    "3/6 !

    #

    '

    7

    #7

    23

    3 "4

    "5

    "3/6 !

    6

    89

    9(6

    '

    7

    #

  • !""

    16. What are the advantages of indirect learning architecture?

    The advantages of indirect learning architecture are:

    The network can be trained on-line, since while one network is being trained, other can perform control function.

    Training can be performed in the region of interest as the input to Neurocontroller is the desired response

    Neural network learns continuously and hence is adaptive. 17. Describe the inverted pendulum control problem?

    The goal of the inverted pendulum task is to apply a sequence of right and left forces of

    fixed magnitude, such that, the pendulum is balanced, and the cart does not hit the edge

    of the track. The cart pendulum physical system is shown in fig.

    18. What are the stages involved in control of inverted pendulum by neural network?

    The control of inverted pendulum by neural network involves following stages:

    Image acquisition Training and Neurocontrol

    :

    09

    9

    #

    ;)9;0:;)9

    *

    09 ;)9

    *0: 9

  • !""

    1. What are the differences between classical (crisp) and fuzzy sets? A crisp set has a unique membership function to represent it, but a fuzzy set can have

    infinite number of membership functions to represent it.

    Crisp set takes only the binary values 0 and 1, whereas a fuzzy set can take values

    between 0 and 1.

    2. What is university of discourse of a set? University of discourse is the universe of all available information on a given problem.

    The University of Discourse is usually denoted by X.

    3. What is cardinal number of a set? The total number of elements in a universe X is called its cardinal number, which is

    denoted by? @ AB. Calculate the cardinal number of X, power set and cardinality of the power set.

    Cardinal number = number of elements in the universe =< 3 Power set, p(x) =>C >DB >(B >EB >D (B >( EB >D EB >D ( EBB. Cardinality of powers set, FM N J# N LB Intersection: JOL >M N JD N LB Complement: JP >M Q J N RB Difference: JML >M N JD Q LB

    8. Write the Exclude middle laws for a classical set. Law of the excluded middle: JKJP R Law of contradiction: JOJP S

    9. What is a fuzzy set? Fuzzy set is a set containing elements that have varying degrees of membership in the set.

    Fuzzy sets are usually denoted by TU . TU maps elements of fuzzy set TU to a real numbered value on the interval 0 to 1.

    10. What are the operations on a fuzzy set? Union: V TW K TX V TW Y V TX Z[\V TW V TX Intersection: V TW O TX V TW ] V TX Z^_V TW V TX Complement: V TW` ` & V TW Difference: V TW a TX V TW O V TX` ` Z^_V TW V TX` `

    11. Write the Exclude middle laws for a fuzzy set.

    Law of the excluded middle: TU K TUb c R Law of contradiction: TU O TUb c S

  • !""

    12. State De Morgans law for a fuzzy set.

    TU O Td` ` ` ` TUb K Tdb D TU K Td` ` ` ` TUb O Tdb 13. What is a fuzzy relation?

    A fuzzy relation, Te is a mapping from the Cartesian space R f g to the interval [0,1], where the strength of the mapping is expressed by the membership function of the

    relation.

    14. What is the difference between a crisp and fuzzy relation? The difference between crisp and fuzzy relation is that the strength of mapping of the

    crisp relation is expressed by the characteristic function, whereas for fuzzy relations, it is

    expressed by the membership function.

    15. Give an expression for cardinality of a crisp relation. What is cardinality of a fuzzy relation?

    The cardinality of a crisp relation is given byG3h3i, where nx and ny are the number of elements (cardinality) of the universe X and Y respectively.

    Since the cardinality of a fuzzy set on any universe is infinity, the cardinality of fuzzy

    relation between two or more universes is also infinity.

    16. State excluded middle law for fuzzy relation.

    Law of the excluded middle: Te K Teb c * j#*kEl1#D Law of contradiction: Te O Teb c mj#mkn#D

    17. What is fuzzy Cartesian product? Let TU be a fuzzy set on universe X and Td be a fuzzy set on universe Y; then the Cartesian product between fuzzy sets TU D Td will result in a fuzzy relation Te , which is contained within the full Cartesian product space.

    To+p+U f Td Te q R f g , where the fuzzy relation Te has a membership function: V Tr V TfX TW lV TW V TX

    18. What is fuzzy max-min composition? Suppose Te is a fuzzy relation on the Cartesian space R f g , Ts is a fuuzy relation on g f t , and T) is a fuzzy relation on R f t ; then fuzzy max-min composition is defined as :

    T) Te u Ts V Tv w Y xV Tr ] V Ty wz{N|

    } V Tv w Z[\Z^_V Tr V Ty w 19. What is fuzzy max-product composition?

    Suppose Te is a fuzzy relation on the Cartesian space R f g , Ts is a fuuzy relation on g f t , and T) is a fuzzy relation on R f t ; then fuzzy max-product composition is defined as :

    V Tv w Y ~V Tr V Ty w{N| } V Tv w Z[\V Tr V Ty w

  • !""

    20. What is a fuzzy equivalence relation? A fuzzy relation Te , on a single universe X is also a relation from X to X. It is a fuzzy equivalence relation if it satisfies the Reflexivity, Symmetry and Transitivity properties.

    Reflexivity: V Tr Symmetry: V Tr x .z V Tr x. z Transitivity: If V Tr x .z D V Tr x. z 4

    Then V Tr j# Z^_ 4

    21. What is a fuzzy tolerance relation? A fuzzy tolerance relation Te on a universe X is a relation that exhibits only the properties of reflexivity and symmetry.

    22. What is fuzzification? Fuzzification is the process of making a crisp quantity fuzzy.

    23. What are the methods of fuzzification? The various methods used for fuzzification are:

    Intuition

    Inference

    Rank ordering

    Angular fuzzy sets

    Neural networks

    Genetic algorithms

    Inductive reasoning

    Soft-partitioning

    Meta rules and

    Fuzzy statics 24. What is defuzzification?

    Defuzzification is the conversion of a fuzzy quantity to a precise (crisp) quantity.

    25. Mention some methods of defuzzification. The methods available for converting a fuzzy quantity to crisp quantity are:

    Max-membership principle

    Centroid method

    Weighted average method

    Mean-max membership method

    Center of sums

    Center of largest area

    First (or last) of maxima 26. What is the drawback of max-membership method of defuzzification?

    Max-membership method, also known as height method is limited to peaked output

    functions.

  • !""

    27. Mention the limitation of weighted average method. This method is valid only for symmetrical output membership functions.

    28. Mention the criteria for selection of a defuzzification method. The criteria for selection of a defuzzification method are:

    It should be continuous There should not be any ambiguity in the obtained crisp value It should be plausible It should be computationally simple.

    1. What is a membership function? Membership function is one that characterizes the fuzziness in a fuzzy set in graphical

    form.

    2. Define core, support and boundary of a membership function. Core is that region of universe that is

    characterized by complete and full

    membership (V TW ). Support is that region of universe

    that is characterized by a non-zero

    membership (V TW ). Boundary is that region of universe

    characterized by non-zero

    membership but not complete

    membership ( V TW )

    3. What is a normal fuzzy set? A normal fuzzy set is one whose membership function has at least one element x in the

    universe whose membership value is unity.

    4. What is a convex fuzzy set? It is a fuzzy set whose membership values are strictly monotonically increasing or

    decreasing or first increasing and then decreasing with increasing values for elements in

    the universe.

    5. What is crossover point of a membership function? Crossover points of a membership function are defined as the elements in the universe for

    which a fuzzy set TU has values equal to 0.5. i.e. V TW +, 6. Define height of a fuzzy set.

    The height of a fuzzy set is the maximum value of membership function.

    7. What are the methods available for fuzzy decision making? Fuzzy synthetic evaluation Fuzzy ordering Fuzzy Bayesian decision method

    x

    =%#&

    >))

    ?9

  • !""

    8. What are the assumptions in fuzzy control system design? The assumptions are:

    The plant is observable and controllable Knowledge base exists Solution to problem exists Engineer is looking for a good enough solution but not for an optimum one Controller is designed to the best of knowledge and acceptable range of precision Stability and optimality is not guaranteed.

    9. What are the features of fuzzy logic control system? The features of a simple fuzzy control system are:

    Fixed and uniform input and output scaling factors Flat, single-partition rule-base with fixed and non-interactive rules Fixed membership functions Limited number of rules, which increases exponentially with number of input

    variables

    Fixed meta knowledge Low level control and no hierarchical rule structure

    10. What is an adaptive fuzzy system? An adaptive fuzzy system is one which adjusts to time or process phased conditions, and

    also changes the supporting system controls such as the rules, the topology of fuzzy sets

    and the method of defuzzification.

    11. What is genetic algorithm? Genetic algorithms are computerized search and optimization algorithms based on

    mechanics of natural genetics and natural selection.

    53@

    1. How is a fuzzy logic controller used to control Blood Pressure during anesthesia? A FLC is used to control the Mean arterial pressure (MAP), which is a measure of the

    depth of anesthesia. The control structure is as shown in fig below.

    @)>*(

    9

    2

    $

    >

    9A

    '

    (9

    A9

    )

    49

    A9

    )

  • !""

    2. What is a neuro-fuzzy controller? It is a combination of neural network and fuzzy system used for control of plant. Neural

    networks are used for optimization and determination of membership functions of fuzzy

    system.

    3. Mention some applications of fuzzy logic. Control of inverted pendulum Aircraft landing control Blood pressure control during anesthesia Image processing Home heating system

    4. How is fuzzy logic used in image processing application? Fuzzy logic is used in image processing in following areas:

    Auto-focus in camera Image stabilization for camcorders Image quality control of TV set

    5. Describe the inverted pendulum control problem? The goal of the inverted pendulum task is to apply a sequence of right and left forces of

    fixed magnitude, such that, the pendulum is balanced, and the cart does not hit the edge

    of the track. The cart pendulum physical system is shown in fig.

    :

    09

    9

    #

    ;)9;0:;)9

    *

    09 ;)9

    *0: 9