Neural Network Pune University MCQs

This set of Neural Networks Multiple Choice Questions & Answers  focuses on “Introduction″.

1. Why do we need biological neural networks?

a) to solve tasks like machine vision & natural language processing

b) to apply heuristic search methods to find solutions of problem

c) to make smart human interactive & user friendly system

d) all of the mentioned

Answer: d

Explanation: These are the basic aims that a neural network achieve.

2. What is the trend in software nowadays?

a) to bring computer more & more closer to user

b) to solve complex problems

c) to be task specific

d) to be versatile

Answer: a

Explanation: Software should be more interactive to the user, so that it can understand its problem in a better fashion.

3. What’s the main point of difference between human & machine intelligence?

a) human perceive everything as a pattern while machine perceive it merely as data

b) human have emotions

c) human have more IQ & intellect

d) human have sense organs

Answer: a

Explanation: Humans have emotions & thus form different patterns on that basis, while a machine is dumb & everything is just a data for him.

4. What is auto-association task in neural networks?

a) find relation between 2 consecutive inputs

b) related to storage & recall task

c) predicting the future inputs

d) none of the mentioned

Answer: b

Explanation: This is the basic definition of auto-association in neural networks.

5. Does pattern classification belongs to category of non-supervised learning?

a) yes

b) no

Answer: b

Explanation: Pattern classification belongs to category of supervised learning.

6. In pattern mapping problem in neural nets, is there any kind of generalization involved between input & output?

a) yes

b) no

Answer: a

Explanation: The desired output is mapped closest to the ideal output & hence there is generalisation involved.

7. What is unsupervised learning?

a) features of group explicitly stated

b) number of groups may be known

c) neither feature & nor number of groups is known

d) none of the mentioned

Answer: c

Explanation: Basic definition of unsupervised learning.

8. Does pattern classification & grouping involve same kind of learning?

a) yes

b) no

Answer: b

Explanation: Pattern classification involves supervised learning while grouping is an unsupervised one.

9. Does for feature mapping there’s need of supervised learning?

a) yes

b) no

Answer: b

Explanation: Feature mapping can be unsupervised, so it’s not a sufficient condition.

10. Example of a unsupervised feature map?

a) text recognition

b) voice recognition

c) image recognition

d) none of the mentioned

Answer: b

Explanation: Since same vowel may occur in different context & its features vary over overlapping regions of different vowels.

11. What is plasticity in neural networks?

a) input pattern keeps on changing

b) input pattern has become static

c) output pattern keeps on changing

d) output is static

Answer: a

Explanation: Dynamic nature of input patterns in an AI problem.

12. What is stability plasticity dilemma ?

a) system can neither be stable nor plastic

b) static inputs & categorization can’t be handled

c) dynamic inputs & categorization can’t be handled

d) none of the mentioned

Answer: c

Explanation: If system is allowed to change its categorization according to inputs it cannot be used for patterns classification & assessment.

13. Drawbacks of template matching are?

a) time consuming

b) highly restricted

c) more generalized

d) none of the the mentioned

Answer: b

Explanation: Point to point pattern matching is carried out in the process.

This set of Neural Networks Multiple Choice Questions & Answers  focuses on “Characteristics – 1″.

1. What are the issues on which biological networks proves to be superior than AI networks?

a) robustness & fault tolerance

b) flexibility

c) collective computation

d) all of the mentioned

Answer: d

Explanation: AI network should be all of the above mentioned.

2. The fundamental unit of network is

a) brain

b) nucleus

c) neuron

d) axon

Answer: c

Explanation: Neuron is the most basic & fundamental unit of a network .

3. What are dendrites?

a) fibers of nerves

b) nuclear projections

c) other name for nucleus

d) none of the mentioned

Answer: a

Explanation: Dendrites tree shaped fibers of nerves.

4. What is shape of dendrites like

a) oval

b) round

c) tree

d) rectangular

Answer: c

Explanation: Basic biological q&a.

5. Signal transmission at synapse is a?

a) physical process

b) chemical process

c) physical & chemical both

d) none of the mentioned

Answer: b

Explanation: Since chemicals are involved at synapse , so its an chemical process.

6. How does the transmission/pulse acknowledged ?

a) by lowering electric potential of neuron body

b) by raising electric potential of neuron body

c) both by lowering & raising electric potential

d) none of the mentioned

Answer: c

Explanation: There is equal probability of both.

7. When the cell is said to be fired?

a) if potential of body reaches a steady threshold values

b) if there is impulse reaction

c) during upbeat of heart

d) none of the mentioned

Answer: a

Explanation: Cell is said to be fired if & only if potential of body reaches a certain steady threshold values.

8. Where does the chemical reactions take place in neuron?

a) dendrites

b) axon

c) synapses

d) nucleus

Answer: c

Explanation: It is a simple biological fact.

9. Function of dendrites is?

a) receptors

b) transmitter

c) both receptor & transmitter

d) none of the mentioned

Answer: a

Explanation: Dendrites are tree like projections whose function is only to receive impulse.

10. What is purpose of Axon?

a) receptors

b) transmitter

c) transmission

d) none of the mentioned

Answer: c

Explanation: Axon is the body of neuron & thus cant be at ends of it so cant receive & transmit signals.

This set of Neural Networks online quiz focuses on “Characteristics – 2”.

1. What is approx size of neuron body?

a) below 5

b) 5-10

c) 10-80

d) above 100

Answer: c

Explanation: Average size of neuron body lies in the above limit.

2. What is the gap at synapses?

a) 50

b) 100

c) 150

d) 200

Answer: d

Explanation: It is near to 200nm.

3. What is charge at protoplasm in state of inactivity?

a) positive

b) negative

c) neutral

d) may be positive or negative

Answer: b

Explanation: It is due to the presence of potassium ion on outer surface in neural fluid.

4. What is the main constituent of neural liquid?

a) sodium

b) potassium

c) Iron

d) none of the mentioned

Answer: a

Explanation: Potassium is the main constituent of neural liquid & responsible for potential on neuron body.

5. What is average potential of neural liquid in inactive state?

a) +70mv

b) +35mv

c) -35mv

d) -70mv

Answer: d

Explanation: It is a basic fact, founded out by series of experiments conducted by neural scientist.

6. At what potential does cell membrane looses it impermeability against Na+ ions?

a) -50mv

b) -35mv

c) -60mv

d) -65mv

Answer: c

Explanation: Cell membrane looses it impermeability against Na+ ions at -60mv.

7. What is effect on neuron as a whole when its potential get raised to -60mv?

a) it get fired

b) no effect

c) it get compressed

d) it expands

Answer: a

Explanation: Cell membrane looses it impermeability against Na+ ions at -60mv.

8. The membrane which allows neural liquid to flow will?

a) never be imperturbable to neural liquid

b) regenerate & retain its original capacity

c) only the certain part get affected, while rest becomes imperturbable again

d) none of the mentioned

Answer: b

Explanation: Each cell of human body has regenerative capacity.

9. How fast is propagation of discharge signal in cells of human brain?

a) less than 0.1m/s

b) 0.5-2m/s

c) 2-5m/s

d) 5-10m/s

Answer: b

Explanation: The process is very fast but comparable to the length of neuron.

10. What is the function of neurotransmitter ?

a) they transmit data directly at synapse to other neuron

b) they modify conductance of post synaptic membrane for certain ions

c) cause polarisation or depolarisation

d) both polarisation & modify conductance of membrane

Answer: d

Explanation: Excitatory & inhibilatory activities are result of these two process.

This set of Neural Networks Questions & Answers for entrance exams focuses on “Characteristics – 3”.

1. The cell body of neuron can be analogous to what mathamatical operation?

a) summing

b) differentiator

c) integrator

d) none of the mentioned

Answer: a

Explanation: Because adding of potential at different parts of neuron is the reason of its firing.

2. What is the critical threshold voltage value at which neuron get fired?

a) 30mv

b) 20mv

c) 25mv

d) 10mv

Answer: d

Explanation: This critical is founded by series of experiments conducted by neural scientist.

3. Does there is any effect on particular neuron which got repeatedly fired ?

a) yes

b) no

Answer: a

Explanation: The strength of neuron to fire in future increases.

4. What is name of above mechanism?

a) hebb rule learning

b) error correction learning

c) memory based learning

d) none of the mentioned

Answer: a

Explanation: It follows from basic definition of hebb rule learning.

5. What is hebb’s rule of learning

a) the system learns from its past mistakes

b) the system recalls previous reference inputs & respective ideal outputs

c) the strength of neural connection get modified accordingly

d) none of the mentioned

Answer:c

Explanation: The strength of neuron to fire in future increases, if it is fired repeatedly.

6. Are all neuron in brain are of same type?

a) yes

b) no

Answer: b

Explanation: Follows from the fact no two body cells are exactly similar in human body, even if they belong to same class.

7. What is estimate number of neurons in human cortex?

a) 10 8

b) 10 5

c) 10 11

d) 10 20

Answer: c

Explanation: It is a fact !

8. what is estimated density of neuron per mm^2 of cortex?

a) 15*(10 2 )

b) 15*(10 4 )

c) 15*(10 3 )

d) 5*(10 4 )

Answer: b

Explanation: It is a biological fact !

9. Why can’t we design a perfect neural network?

a) full operation is still not known of biological neurons

b) number of neuron is itself not precisely known

c) number of interconnection is very large & is very complex

d) all of the mentioned

Answer: d

Explanation: These are all fundamental reasons, why can’t we design a perfect neural network !

10. How many synaptic connection are there in human brain?

a) 10 10

b) 10 15

c) 10 20

d) 10 5

Answer: b

Explanation: You can estimate this value from number of neurons in human cortex & their density.

This set of Neural Networks Multiple Choice Questions & Answers  focuses on “History″.

1. Operations in the neural networks can perform what kind of operations?

a) serial

b) parallel

c) serial or parallel

d) none of the mentioned

Answer: c

Explanation: General characteristics of neural networks.

2. Does the argument information in brain is adaptable, whereas in the computer it is replaceable is valid?

a) yes

b) no

Answer: a

Explanation: Its a fact & related to basic knowledge of neural networks !

3. Does there exist central control for processing information in brain as in computer?

a) yes

b) no

Answer: b

Explanation: In human brain information is locally processed & analysed.

4. Which action is faster pattern classification or adjustment of weights in neural nets?

a) pattern classification

b) adjustment of weights

c) equal

d) either of them can be fast, depending on conditions

Answer: a

Explanation: Memory is addressable, so thus pattern can be easily classified.

5. What is the feature of ANNs due to which they can deal with noisy, fuzzy, inconsistent data?

a) associative nature of networks

b) distributive nature of networks

c) both associative & distributive

d) none of the mentioned

Answer: c

Explanation: General characteristics of ANNs.

6. What was the name of the first model which can perform wieghted sum of inputs?

a) McCulloch-pitts neuron model

b) Marvin Minsky neuron model

c) Hopfield model of neuron

d) none of the mentioned

Answer: a

Explanation: McCulloch-pitts neuron model can perform weighted sum of inputs followed by threshold logic operation.

7. Who developed the first learning machine in which connection strengths could be adapted automatically?

a) McCulloch-pitts

b) Marvin Minsky

c) Hopfield

d) none of the mentioned

Answer: b

Explanation: In 1954 Marvin Minsky developed the first learning machine in which connection strengths could be adapted automatically & efficiebtly.

8. Who proposed the first perceptron model in 1958?

a) McCulloch-pitts

b) Marvin Minsky

c) Hopfield

d) Rosenblatt

Answer: d

Explanation: Rosenblatt proposed the first perceptron model in 1958 .

9. John hopfield was credited for what important aspec of neuron?

a) learning algorithms

b) adaptive signal processing

c) energy analysis

d) none of the mentioned

Answer: c

Explanation: It was of major contribution of his works in 1982.

Answer: b

Explanation: Ackley, Hinton built the boltzman machine.

This set of Neural Networks Questions & Answers for campus interviews focuses on “Terminology”.

1. What is ART in neural networks?

a) automatic resonance theory

b) artificial resonance theory

c) adaptive resonance theory

d) none of the mentioned

Answer: c

Explanation: It is full form of ART & is basic q&a.

2. What is an activation value?

a) weighted sum of inputs

b) threshold value

c) main input to neuron

d) none of the mentioned

Answer: a

Explanation: It is definition of activation value & is basic q&a.

3. Positive sign of weight indicates?

a) excitatory input

b) inhibitory input

c) can be either excitatory or inhibitory as such

d) none of the mentioned

Answer: a

Explanation: Sign convention of neuron.

4. Negative sign of weight indicates?

a) excitatory input

b) inhibitory input

c) excitatory output

d) inhibitory output

Answer: b

Explanation: Sign convention of neuron.

5. The amount of output of one unit received by another unit depends on what?

a) output unit

b) input unit

c) activation value

d) weight

Answer: d

Explanation: Activation is sum of wieghted sum of inputs, which gives desired output..hence output depends on weights.

6. The process of adjusting the weight is known as?

a) activation

b) synchronisation

c) learning

d) none of the mentioned

Answer: c

Explanation: Basic definition of learning in neural nets .

7. The procedure to incrementally update each of weights in neural is referred to as?

a) synchronisation

b) learning law

c) learning algorithm

d) both learning algorithm & law

Answer: d

Explanation: Basic definition of learning law in neural.

8. In what ways can output be determined from activation value?

a) deterministically

b) stochastically

c) both deterministically & stochastically

d) none of the mentioned

Answer: c

Explanation: This is the most important trait of input processing & output determination in neural networks.

9. How can output be updated in neural network?

a) synchronously

b) asynchronously

c) both synchronously & asynchronously

d) none of the mentioned

Answer: c

Explanation: Output can be updated at same time or at different time in the networks.

10. What is asynchronous update in neural netwks?

a) output units are updated sequentially

b) output units are updated in parallel fashion

c) can be either sequentially or in parallel fashion

d) none of the mentioned

Answer: a

Explanation: Output are updated at different time in the networks.

This set of Neural Networks Multiple Choice Questions & Answers  focuses on “Models – 1″.

1. What is the name of the model in figure below?

neural-networks-questions-answers-models-1-q1

a) Rosenblatt perceptron model

b) McCulloch-pitts model

c) Widrow’s Adaline model

d) None of the mentioned

Answer: b

Explanation: It is a general block diagram of McCulloch-pitts model of neuron.

2. What is nature of function F in the figure?

a) linear

b) non-linear

c) can be either linear or non-linear

d) none of the mentioned

Answer: b

Explanation: In this function, the independent variable is an exponent in the equation hence non-linear.

3. What does the character ‘b’ represents in the above diagram?

a) bias

b) any constant value

c) a variable value

d) none of the mentioned

Answer: a

Explanation: More appropriate choice since bias is a constant fixed value for any circuit model.

4. If ‘b’ in the figure below is the bias, then what logic circuit does it represents?

neural-networks-questions-answers-models-1-q4

a) or gate

b) and gate

c) nor gate

d) nand gate

Answer: c

Explanation: Form the truth table of above figure by taking inputs as 0 or 1.

5. When both inputs are 1, what will be the output of the above figure?

a) 0

b) 1

c) either 0 or 1

d) z

Answer: a

Explanation: Check the truth table of nor gate.

6. When both inputs are different, what will be the output of the above figure?

a) 0

b) 1

c) either 0 or 1

d) z

Answer: a

Explanation: Check the truth table of nor gate.

7. Which of the following model has ability to learn?

a) pitts model

b) rosenblatt perceptron model

c) both rosenblatt and pitts model

d) neither rosenblatt nor pitts

Answer: b

Explanation: Weights are fixed in pitts model but adjustable in rosenblatt.

8. When both inputs are 1, what will be the output of the pitts model nand gate ?

a) 0

b) 1

c) either 0 or 1

d) z

Answer: a

Explanation: Check the truth table of simply a nand gate.

9. When both inputs are different, what will be the logical output of the figure of question 4?

a) 0

b) 1

c) either 0 or 1

d) z

Answer: a

Explanation: Check the truth table of nor gate.

10. Does McCulloch-pitts model have ability of learning?

a) yes

b) no

Answer: b

Explanation: Weights are fixed.

This set of Neural Networks Interview Questions and Answers focuses on “Models – 2”

1. Who invented perceptron neural networks?

a) McCullocch-pitts

b) Widrow

c) Minsky & papert

d) Rosenblatt

Answer: d

Explanation: The perceptron is one of the earliest neural networks. Invented at the Cornell Aeronautical Laboratory in 1957 by Frank Rosenblatt, the Perceptron was an attempt to understand human memory, learning, and cognitive processes.

2. What was the 2nd stage in perceptron model called?

a) sensory units

b) summing unit

c) association unit

d) output unit

Answer: c

Explanation: This was the very speciality of the perceptron model, that is performs association mapping on outputs of he sensory units.

3. What was the main deviation in perceptron model from that of MP model?

a) more inputs can be incorporated

b) learning enabled

c) all of the mentioned

d) none of the mentioned

Answer: b

Explanation: The weights in perceprton model are adjustable.

4. What is delta  in perceptron model of neuron?

a) error due to environmental condition

b) difference between desired & target output

c) can be both due to difference in target output or environmental condition

d) none of the mentioned

Answer: a

Explanation: All other parameters are assumed to be null while calculatin the error in perceptron model & only difference between desired & target output is taken into account.

5. If a is the input, ^ is the error, n is the learning parameter, then how can weight change in a perceptron model be represented?

a) na

b) n^

c) ^a

d) none of the mentioned

Answer: d

Explanation: The correct answer is n^a.

6. What is adaline in neural networks?

a) adaptive linear element

b) automatic linear element

c) adaptive line element

d) none of the mentioned

Answer: a

Explanation: adaptive linear element is the full form of adaline neural model.

7. who invented the adaline neural model?

a) Rosenblatt

b) Hopfield

c) Werbos

d) Widrow

Answer: d

Explanation: Widrow invented the adaline neural model.

8. What was the main point of difference between the adaline & perceptron model?

a) weights are compared with output

b) sensory units result is compared with output

c) analog activation value is compared with output

d) all of the mentioned

Answer: c

Explanation: Analog activation value comparison with output,instead of desired output as in perceptron model was the main point of difference between the adaline & perceptron model.

9. In adaline model what is the relation between output & activation value?

a) linear

b) nonlinear

c) can be either linear or non-linear

d) none of the mentioned

Answer: a

Explanation: s,output=f=x. Hence its a linear model.

10. what is the another name of weight update rule in adaline model based on its functionality?

a) LMS error learning law

b) gradient descent algorithm

c) both LMS error & gradient descent learning law

d) none of the mentioned

Answer: c

Explanation: weight update rule minimizes the mean squared error, averaged over all inputs & this laws is derived using negative gradient of error surface weight space, hence option a & b.

This set of Neural Networks Multiple Choice Questions & Answers  focuses on “Topology″.

1. In neural how can connectons between different layers be achieved?

a) interlayer

b) intralayer

c) both interlayer and intralayer

d) either interlayer or intralayer

Answer: c

Explanation: Connections between layers can be made to one unit to another and within the units of a layer.

2. Connections across the layers in standard topologies & among the units within a layer can be organised?

a) in feedforward manner

b) in feedback manner

c) both feedforward & feedback

d) either feedforward & feedback

Answer: d

Explanation: Connections across the layers in standard topologies can be in feedforward manner or in feedback manner but not both.

3. What is an instar topology?

a) when input is given to layer F1, the the jth unit of other layer F2 will be activated to maximum extent

b) when weight vector for connections from jth unit  in F2 approaches the activity pattern in F1

c) can be either way

d) none of the mentioned

Answer: a

Explanation: Restatement of basic definition of instar.

4. What is an outstar topology?

a) when input is given to layer F1, the the jth unit of other layer F2 will be activated to maximum extent

b) when weight vector for connections from jth unit  in F2 approaches the activity pattern in F1

c) can be either way

d) none of the mentioned

Answer: b

Explanation: Restatement of basic definition of outstar.

5. The operation of instar can be viewed as?

a) content addressing the memory

b) memory addressing the content

c) either content addressing or memory addressing

d) both content & memory addressing

Answer: a

Explanation: Because in instar, when input is given to layer F1, the the jth unit of other layer F2 will be activated to maximum extent.

6. The operation of outstar can be viewed as?

a) content addressing the memory

b) memory addressing the content

c) either content addressing or memory addressing

d) both content & memory addressing

Answer: b

Explanation: Because in outstar, when weight vector for connections from jth unit  in F2 approaches the activity pattern in F1.

7. If two layers coincide & weights are symmetric, then what is that structure called?

a) instar

b) outstar

c) autoassociative memory

d) heteroassociative memory

Answer: c

Explanation: In autoassociative memory each unit is connected to every other unit & to itself.

8. Heteroassociative memory can be an example of which type of network?

a) group of instars

b) group of oustar

c) either group of instars or outstars

d) both group of instars or outstars

Answer: c

Explanation: Depending upon the flow, the memory can be of either of the type.

9. What is STM in neural network?

a) short topology memory

b) stimulated topology memory

c) short term memory

d) none of the mentioned

Answer: c

Explanation: Full form of STM.

10. What does STM corresponds to?

a) activation state of network

b) encoded pattern information pattern in synaptic weights

c) either way

d) both way

Answer: a

Explanation: Short-term memory  refers to the capacity-limited retention of information over a brief period of time,hence the option.

11. What LTM corresponds to?

a) activation state of network

b) encoded pattern information pattern in synaptic weights

c) either way

d) both way

Answer: b

Explanation: Long-term memory  & hence the option.

This set of Neural Networks Multiple Choice Questions & Answers  focuses on “Learning – 1″.

1. On what parameters can change in weight vector depend?

a) learning parameters

b) input vector

c) learning signal

d) all of the mentioned

Answer: d

Explanation: Change in weight vector corresponding to jth input at time  depends on all of these parameters.

2. If the change in weight vector is represented by ∆wij, what does it mean?

a) describes the change in weight vector for ith processing unit, taking input vector jth into account

b) describes the change in weight vector for jth processing unit, taking input vector ith into account

c) describes the change in weight vector for jth & ith processing unit.

d) none of the mentioned

Answer: a

Explanation: ∆wij= µfaj, where a is the input vector.

3. What is learning signal in this equation ∆wij= µfaj?

a) µ

b) wi a

c) aj

d) f

Answer: d

Explanation: This the non linear representation of output of the network.

4. State whether Hebb’s law is supervised learning or of unsupervised type?

a) supervised

b) unsupervised

c) either supervised or unsupervised

d) can be both supervised & unsupervised

Answer: b

Explanation: No desired output is required for it’s implementation.

5. Hebb’s law can be represented by equation?

a) ∆wij= µfaj

b) ∆wij= µ aj, where  is output signal of ith input

c) both way

d) none of the mentioned

Answer: c

Explanation: = f, in Hebb’s law.

6. State which of the following statements hold foe perceptron learning law?

a) it is supervised type of learning law

b) it requires desired output for each input

c) ∆wij= µ aj

d) all of the mentioned

Answer: d

Explanation: all statements follow from ∆wij= µ aj, where bi is the target output & hence supervised learning.

7. Delta learning is of unsupervised type?

a) yes

b) no

Answer: b

Explanation: Change in weight is based on the error between the desired & the actual output values for a given input.

8. widrow & hoff learning law is special case of?

a) hebb learning law

b) perceptron learning law

c) delta learning law

d) none of the mentioned

Answer: c

Explanation: Output function in this law is assumed to be linear , all other things same.

9. What’s the other name of widrow & hoff learning law?

a) Hebb

b) LMS

c) MMS

d) None of the mentioned

Answer: b

Explanation: LMS, least mean square. Change in weight is made proportional to negative gradient of error & due to linearity of output function.

10. Which of the following equation represent perceptron learning law?

a) ∆wij= µ aj

b) ∆wij= µ aj

c) ∆wij= µ aj Á,wher Á is derivative of xi

d) ∆wij= µ) aj

Answer: b

Explanation: Perceptron learning law is supervised, nonlinear type of learning.

This set of Neural Networks Multiple Choice Questions and Answers for freshers focuses on “Learning – 2”.

1. Correlation learning law is special case of?

a) Hebb learning law

b) Perceptron learning law

c) Delta learning law

d) LMS learning law

Answer: a

Explanation: Since in hebb is replaced by bi in correlation.

2. Correlation learning law is what type of learning?

a) supervised

b) unsupervised

c) either supervised or unsupervised

d) both supervised or unsupervised

Answer: a

Explanation: Supervised, since depends on target output.

3. Correlation learning law can be represented by equation?

a) ∆wij= µ aj

b) ∆wij= µ aj

c) ∆wij= µ aj Á,where Á is derivative of xi

d) ∆wij= µ bi aj

Answer: d

Explanation: Correlation learning law depends on target output.

4. The other name for instar learning law?

a) looser take it all

b) winner take it all

c) winner give it all

d) looser give it all

Answer: b

Explanation: The unit which gives maximum output, weight is adjusted for that unit.

5. The instar learning law can be represented by equation?

a) ∆wij= µ aj

b) ∆wij= µ aj

c) ∆wij= µ aj Á,where Á is derivative of xi

d) ∆wk= µ , unit k with maximum output is identified

Answer: d

Explanation: Follows from basic definition of instar learning law.

6. Is instar a case of supervised learning?

a) yes

b) no

Answer: b

Explanation: Since weight adjustment don’t depend on target output, it is unsupervised learning.

7. The instar learning law can be represented by equation?

a) ∆wjk= µ, where the kth unit is the only active in the input layer

b) ∆wij= µ aj

c) ∆wij= µ aj Á,wher Á is derivative of xi

d) ∆wij= µ aj

Answer: a

Explanation: Follows from basic definition of outstar learning law.

8. Is outstar a case of supervised learning?

a) yes

b) no

Answer: a

Explanation: Since weight adjustment depend on target output, it is supervised learning.

9. Which of the following learning laws belongs to same category of learning?

a) hebbian, perceptron

b) perceptron, delta

c) hebbian, widrow-hoff

d) instar, outstar

Answer: b

Explanation: They both belongs to supervised type learning.

10. In hebbian learning intial weights are set?

a) random

b) near to zero

c) near to target value

d) near to target value

Answer: b

Explanation: Hebb law lead to sum of correlations between input & output, inorder to achieve this, the starting initial weight values must be small.

This set of Neural Networks Multiple Choice Questions & Answers  focuses on “Dynamics″.

1. Weight state i.e set of weight values are determined by what kind of dynamics?

a) synaptic dynamics

b) neural level dynamics

c) can be either synaptic or neural dynamics

d) none of the mentioned

Answer: a

Explanation: Weights are best determined by synaptic dynamics, as it is one fastest & precise dynamics occurring.

2. Which is faster neural level dynamics or synaptic dynamics?

a) neural level

b) synaptic

c) both equal

d) insufficient information

Answer: a

Explanation: Since neural level dyna,ics depends on input fluctuations & these take place at every milliseconds.

3. During activation dynamics does weight changes?

a) yes

b) no

Answer: b

Explanation: During activation dynamics, synaptic weights don’t change significantly & hence assumed to be constant.

4. Activation dynamics is referred as?

a) short term memory

b) long term memory

c) either short or long term

d) both short & long term

Answer: a

Explanation: It depends on input pattern, & input changes from moment to moment, hence Short term memory.

5. Synaptic dynamics is referred as?

a) short term memory

b) long term memory

c) either short or long term

d) both short & long term

Answer: b

Explanation: Synaptic dynamics don’t change for a given set of training inputs, hence long term memory.

6. What is classification?

a) deciding what features to use in a pattern recognition problem

b) deciding what class an input pattern belongs to

c) deciding what type of neural network to use

d) none of the mentioned

Answer: b

Explanation: Follows from basic definition of classification.

7. What is generalization?

a) the ability of a pattern recognition system to approximate the desired output values for pattern vectors which are not in the test set.

b) the ability of a pattern recognition system to approximate the desired output values for pattern vectors which are not in the training set.

c) can be either way

d) none of the mentioned

Answer: b

Explanation: Follows from basic definition of generalization.

8. What are models in neural networks?

a) mathematical representation of our understanding

b) representation of biological neural networks

c) both way

d) none of the mentioned

Answer: c

Explanation: Model should be close to our biological neural systems, so that we can have high efficiency in machines too.

9. What kind of dynamics leads to learning laws?

a) synaptic

b) neural

c) activation

d) both synaptic & neural

Answer: a

Explanation: Since weights are dependent on synaptic dynamics, hence learning laws.

Answer: c

Explanation: Activation dynamics depends on input pattern, hence any change in input pattern will affect activation dynamics of neural networks.

Participate in the Sanfoundry Certification contest to get free Certificate of Merit. Join our social networks below and stay updated with latest contests, videos, internships and jobs!

This set of Neural Networks Multiple Choice Questions & Answers  focuses on “Activation Models″.

1. Activation value is associated with?

a) potential at synapses

b) cell membrane potential

c) all of the mentioned

d) none of the mentioned

Answer: b

Explanation: Cell membrane potential determines the activation value in neural nets.

2. In activation dynamics is output function bounded?

a) yes

b) no

Answer: a

Explanation: It is the nature of output function in activation dynamics.

3. What’s the actual reason behind the boundedness of the output function in activation dynamics?

a) limited neural fluid

b) limited fan in capacity of inputs

c) both limited neural fluid & fan in capacity

d) none of the mentioned

Answer: d

Explanation: It is due to the limited current carrying capacity of cell membrane.

4. What is noise saturation dilemma?

a) at saturation state neuron will stop working, while biologically it’s not feasible

b) how can a neuron with limited operating range be made sensitive to nearly unlimited range of inputs

c) can be either way

d) none of the mentioned

Answer: b

Explanation: Threshold value setting has to be adjusted properly.

5. Broadly how many kinds of stability can be defined in neural networks?

a) 1

b) 3

c) 2

d) 4

Answer: c

Explanation: There exist broadly structural & global stability in neural.

6. What is structural stability?

a) when both synaptic & activation dynamics are simultaneously used & are in equilibrium

b) when only synaptic dynamics in equilibrium

c) when only synaptic dynamics in equilibrium

d) none of the mentioned

Answer: d

Explanation: Refers to state equilibrium situation where small perturbations brings network back to equilibrium.

7. What is global stability?

a) when both synaptic & activation dynamics are simultaneously used & are in equilibrium

b) when only synaptic & activation dynamics are used

c) when only synaptic dynamics in equilibrium

d) none of the mentioned

Answer: a

Explanation: Global stability means neuron as a whole is stable.

8. Which models belongs to main subcategory of activation models?

a) additive & subtractive activation models

b) additive & shunting activation models

c) subtractive & shunting activation models

d) all of the mentioned

Answer: b

Explanation: Additive & shunting activation models are the most basic category of activation models.

9.What is the assumption of perkels model, if f is the output function in additive activation model?

a) f=x

b) f=x 2

c) f=x 3

d) none of the mentioned

Answer: a

Explanation: Perkels model assumes output function to be linear.

10. Who proposed the shunting activation model?

a) rosenblatt

b) hopfield

c) perkel

d) grossberg

Answer: d

Explanation: Grossberg proposed the model in 1982.

11. What was the goal of shunting activation model?

a) to make system dynamic

b) to keep operating range of activation value to a specified range

c) to make system static

d) can be either for dynamic or static, depending on inputs

Answer: b

Explanation: Stabilizing & bounding the unbounded range of activation value was the primary goal of this model.

This set of Neural Networks Aptitude Test focuses on “Learning Basics – 1”.

1. Activation models are?

a) dynamic

b) static

c) deterministic

d) none of the mentioned

Answer: c

Explanation: Input/output patterns & the activation values may be considered as sample functions of random process.

2. If xb represents differentiation of state x, then a stochastic model can be represented by?

a) xb=deterministic model

b) xb=deterministic model + noise component

c) xb=deterministic model*noise component

d) none of the mentioned’

Answer: b

Explanation: Noise is assumed to be additive in nature in stochastic models.

3. What is equilibrium in neural systems?

a) deviation in present state, when small perturbations occur

b) settlement of network, when small perturbations occur

c) change in state, when small perturbations occur

d) none of the mentioned

Answer: b

Explanation: Follows from basic definition of equilibrium.

4.What is the condition in Stochastic models, if xb represents differentiation of state x?

a) xb=0

b) xb=1

c) xb=n, where n is noise component

d) xb=n+1

Answer: c

Explanation: xb=0 is condition for deterministic models, so option c is radical choice.

5. What is asynchronous update in a network?

a) update to all units is done at the same time

b) change in state of any one unit drive the whole network

c) change in state of any number of units drive the whole network

d) none of the mentioned

Answer: b

Explanation: In asynchronous update, change in state of any one unit drive the whole network.

6. Learning is a?

a) slow process

b) fast process

c) can be slow or fast in general

d) can’t say

Answer: a

Explanation: Learning is a slow process.

7. What are the requirements of learning laws?

a) convergence of weights

b) learning time should be as small as possible

c) learning should use only local weights

d) all of the mentioned

Answer: d

Explanation: These all are the some of basic requirements of learning laws.

8. Memory decay affects what kind of memory?

a) short tem memory in general

b) older memory in general

c) can be short term or older

d) none of the mentioned

Answer: a

Explanation: Memory decay affects short term memory rather than older memories.

9. What are the requirements of learning laws?

a) learning should be able to capture more & more patterns

b) learning should be able to grasp complex nonliear mappings

c) convergence of weights

d) all of the mentioned

Answer: d

Explanation: These all are the some of basic requirements of learning laws.

10. How is pattern information distributed?

a) it is distributed all across the weights

b) it is distributed in localised weights

c) it is distributed in certain proctive weights only

d) none of the mentioned

Answer: a

Explanation: pattern information is highly distributed all across the weights.

This set of Neural Networks Inteview Questions and Answers for freshers focuses on “Learning Basics – 2”.

1. What is supervised learning?

a) weight adjustment based on deviation of desired output from actual output

b) weight adjustment based on desired output only

c) weight adjustment based on actual output only

d) none of the mentioned

Answer: a

Explanation: Supervised learning is based on weight adjustment based on deviation of desired output from actual output.

2. Supervised learning may be used for?

a) temporal learning

b) structural learning

c) both temporal & structural learning

d) none of the mentioned

Answer: c

Explanation: Supervised learning may be used for both temporal & structural learning.

3. What is structural learning?

a) concerned with capturing input-output relationship in patterns

b) concerned with capturing weight relationships

c) both weight & input-output relationships

d) none of the mentioned

Answer: a

Explanation: Structural learning deals with learning the overall structure of network in a macroscopic view.

4. What is temporal learning?

a) concerned with capturing input-output relationship in patterns

b) concerned with capturing weight relationships

c) both weight & input-output relationships

d) none of the mentioned

Answer: b

Explanation: Temporal learning is concerned with capturing weight relationships.

5. What is unsupervised learning?

a) weight adjustment based on deviation of desired output from actual output

b) weight adjustment based on desired output only

c) weight adjustment based on local information available to weights

d) none of the mentioned

Answer: c

Explanation: Unsupervised learning is purely based on adjustment based on local information available to weights.

6. Learning methods can only be online?

a) yes

b) no

Answer: b

Explanation: Learning can be offline too.

7. Online learning allows network to incrementally adjust weights continuously?

a) yes

b) no

Answer: a

Explanation: Follows from basic definition of online learning.

8. What is nature of input in activation dynamics?

a) static

b) dynamic

c) both static & dynamic

d) none of the mentioned

Answer: a

Explanation: Input is fixed throughout the dynamics.

9. Adjustments in activation is slower than that of synaptic weights?

a) yes

b) no

Answer: b

Explanation: Adjustments in activation is faster than that of synaptic weights.

10. what does the term wij represents in synaptic dynamic model?

a) a prioi knowledge

b) just a constant

c) no strong significance

d) future adjustments

Answer: a

Explanation: Refer to weight equation of synaptic dynamic model.

This set of Neural Networks Multiple Choice Questions & Answers  focuses on “Learning Laws-1″.

1. What is hebbian learning?

a) synaptic strength is proportional to correlation between firing of post & presynaptic neuron

b) synaptic strength is proportional to correlation between firing of postsynaptic neuron only

c) synaptic strength is proportional to correlation between firing of presynaptic neuron only

d) none of the mentioned

Answer: a

Explanation: Folllows from basic definition of hebbian learning.

2. What is differential hebbian learning?

a) synaptic strength is proportional to correlation between firing of post & presynaptic neuron

b) synaptic strength is proportional to correlation between firing of postsynaptic neuron only

c) synaptic strength is proportional to correlation between firing of presynaptic neuron only

d) synaptic strength is proportional to changes in correlation between firing of post & presynaptic neuron

Answer: d

Explanation: Differential hebbian learning is proportional to changes in correlation between firing of post & presynaptic neuron.

3. What is competitive learning?

a) learning laws which modulate difference between synaptic weight & output signal

b) learning laws which modulate difference between synaptic weight & activation value

c) learning laws which modulate difference between actual output & desired output

d) none of the mentioned

Answer: a

Explanation: Competitive learning laws modulate difference between synaptic weight & output signal.

4. What is differential competitive learning?

a) synaptic strength is proportional to changes of post & presynaptic neuron

b) synaptic strength is proportional to changes of postsynaptic neuron only

c) synaptic strength is proportional to changes of presynaptic neuron only

d) none of the mentioned

Answer: d

Explanation: Differential competitive learning is based on to changes of postsynaptic neuron only.

5. What is error correction learning?

a) learning laws which modulate difference between synaptic weight & output signal

b) learning laws which modulate difference between synaptic weight & activation value

c) learning laws which modulate difference between actual output & desired output

d) none of the mentioned

Answer: c

Explanation: Error correction learning is base on difference between actual output & desired output.

6. Continuous perceptron learning is also known as delta learning?

a) yes

b) no

Answer: a

Explanation: Follows from basic definition of delta learning.

7. Widrows LMS algorithm is also based on error correction learning?

a) yes

b) no

Answer: a

Explanation: It uses the instantaneous squared error between desired & actual output of unit.

8. Error correction learning is type of?

a) supervised learning

b) unsupervised learning

c) can be supervised or unsupervised

d) none of the mentioned

Answer: a

Explanation: Since desired output for an input is known.

9. Error correction learning is like learning with teacher?

a) yes

b) no

Answer: a

Explanation: Since desired output for an input is known.

10. What is reinforcement learning?

a) learning is based on evaluative signal

b) learning is based o desired output for an input

c) learning is based on both desired output & evaluative signal

d) none of the mentioned

Answer: a

Explanation: Reinforcement learning is based on evaluative signal.

This set of Neural Networks Questions and Answers for experienced focuses on “Learning Laws – 2”.

1. Reinforcement learning is also known as learning with critic?

a) yes

b) no

Answer: a

Explanation: Since this is evaluative & not instructive.

2. How many types of reinforcement learning exist?

a) 2

b) 3

c) 4

d) 5

Answer: b

Explanation: Fixed credit assignment, probablistic credit assignment, temporal credit assignment.

3. What is fixed credit assignment?

a) reinforcement signal given to input-output pair don’t change with time

b) input-output pair determine probability of postive reinforcement

c) input pattern depends on past history

d) none of the mentioned

Answer: a

Explanation: In fixed credit assignment, reinforcement signal given to input-output pair don’t change with time.

4. What is probablistic credit assignment?

a) reinforcement signal given to input-output pair don’t change with time

b) input-output pair determine probability of postive reinforcement

c) input pattern depends on past history

d) none of the mentioned

Answer: b

Explanation: In probablistic credit assignment, input-output pair determine probability of postive reinforcement.

5. What is temporal credit assignment?

a) reinforcement signal given to input-output pair don’t change with time

b) input-output pair determine probability of postive reinforcement

c) input pattern depends on past history

d) none of the mentioned

Answer: c

Explanation: In temporal credit assignment, input pattern depends on past history.

6. Boltzman learning uses what kind of learning?

a) deterministic

b) stochastic

c) either deterministic or stochastic

d) none of the mentioned

Answer: b

Explanation: Boltzman learning uses deterministic learning.

7. Whats true for sparse encoding learning?

a) logical And & Or operations are used for input output relations

b) weight corresponds to minimum & maximum of units are connected

c) weights are expressed as linear combination of orthogonal basis vectors

d) change in weight uses a weighted sum of changes in past input values

Answer: a

Explanation: sparse encoding learning employs Logical And & Or operations are used for input output relations.

8. Whats true for Drive reinforcement learning?

a) logical And & Or operations are used for input output relations

b) weight corresponds to minimum & maximum of units are connected

c) weights are expressed as linear combination of orthogonal basis vectors

d) change in weight uses a weighted sum of changes in past input values

Answer: d

Explanation: In Drive reinforcement learning, change in weight uses a weighted sum of changes in past input values.

9. Whats true for Min-max learning?

a) logical And & Or operations are used for input output relations

b) weight corresponds to minimum & maximum of units are connected

c) weights are expressed as linear combination of orthogonal basis vectors

d) change in weight uses a weighted sum of changes in past input values

Answer: b

Explanation: Min-max learning involves weights which corresponds to minimum & maximum of units connected.

10. Whats true for principal component learning?

a) logical And & Or operations are used for input output relations

b) weight corresponds to minimum & maximum of units are connected

c) weights are expressed as linear combination of orthogonal basis vectors

d) change in weight uses a weighted sum of changes in past input values

Answer: c

Explanation: principal component learning involves weights that are expressed as linear combination of orthogonal basis vectors.

This set of Neural Networks Multiple Choice Questions & Answers  focuses on “Convergence & stability″.

1. Stability refers to adjustment in behaviour of weights during learning?

a) yes

b) no

Answer: b

Explanation: Stability refers to equilibrium behaviour of activation state.

2. Convergence refers to equilibrium behaviour of activation state?

a) yes

b) no

Answer: b

Explanation: Convergence refers to adjustment in behaviour of weights during learning.

3. What leads to minimization of error between the desired & actual outputs?

a) stability

b) convergence

c) either stability or convergence

d) none of the mentioned

Answer: b

Explanation: Convergence is responsible for minimization of error between the desired & actual outputs.

4. Stability is minimization of error between the desired & actual outputs?

a) yes

b) no

Answer: b

Explanation: Convergence is minimization of error between the desired & actual outputs.

5. How many trajectories may terminate at same equilibrium state?

a) 1

c) 2

c) many

d) none

Answer: c

Explanation: There may be several trajectories that may settle to same equilibrium state.

6. If weights are not symmetric i.e cik =! cki, then what happens?

a) network may exhibit periodic oscillations of states

b) no oscillations as it doesn’t depend on it

c) system is stable

d) system in practical equilibrium

Answer: a

Explanation: At this situation system exhibits some unwanted oscillations.

7. Is pattern storage possible if system has chaotic stability?

a) yes

b) no

Answer: a

Explanation: Pattern storage is possible if any network exhibits either fixed point, oscillatory, chaotic stability.

8. If states of system experience basins of attraction, then system may achieve what kind of stability?

a) fixed point stability

b) oscillatory stability

c) chaotic stability

d) none of the mentioned

Answer: c

Explanation: Basins of attraction is a property of chaotic stability.

9. What is an objective of a learning law?

a) to capture pattern information in training set data

b) to modify weights so as to achieve output close to desired output

c) it should lead to convergence of system or its weights

d) all of the mentioned

Answer: d

Explanation: These all are some objectives of learning laws.

10. A network will be useful only if, it leads to equilibrium state at which there is no change of state?

a) yes

b) no

Answer: a

Explanation: Its the basic condition for stability.

This set of Neural Networks Multiple Choice Questions & Answers  focuses on “Recall″.

1. Lyapunov function is vector in nature?

a) yes

b) no

Answer: b

Explanation: Lyapunov function is scalar in nature.

2. What’s the role of lyaopunov fuction?

a) to determine stability

b) to determine convergence

c) both stability & convergence

d) none of the mentioned

Answer: a

Explanation: lyapunov is an energy function.

3. Did existence of lyapunov function is necessary for stability?

a) yes

b) no

Answer: b

Explanation: It is sufficient but not necessary condition.

4. V is said to be lyapunov function if?

a) v >=0

b) v <=0

c) v =0

d) none of the mentioned

Answer: b

Explanation: It is the condition for existence for lyapunov function.

5. What does cohen grossberg theorem?

a) shows the stability of fixed weight autoassociative networks

b) shows the stability of adaptive autoaassociative networks

c) shows the stability of adaptive heteroassociative networks

d) none of the mentioned

Answer: a

Explanation: Cohen grossberg theorem shows the stability of fixed weight autoassociative networks.

6. What does cohen grossberg kosko theorem?

a) shows the stability of fixed weight autoassociative networks

b) shows the stability of adaptive autoaassociative networks

c) shows the stability of adaptive heteroassociative networks

d) none of the mentioned

Answer: b

Explanation: Cohen grossberg kosko shows the stability of adaptive autoaassociative networks.

7. What does 3rd theorem that describe the stability of a set of nonlinear dynamical systems?

a) shows the stability of fixed weight autoassociative networks

b) shows the stability of adaptive autoaassociative networks

c) shows the stability of adaptive heteroassociative networks

d) none of the mentioned

Answer: c

Explanation: 3rd theorem of nonlinear dynamical systems, shows the stability of adaptive heteroassociative networks.

8. What happens during recall in neural networks?

a) weight changes are suppressed

b) input to the network determines the output activation

c) both process has to happen

d) none of the mentioned

Answer: c

Explanation: Follows from basic definition of Recall in a network.

9. Can a neural network learn & recall at the same time?

a) yes

b) no

Answer: a

Explanation: It was later proved by kosko in 1988.

10. In nearest neighbour case, the stored pattern closest to input pattern is recalled, where does it occurs?

a) feedback pattern classification

b) feedforward pattern classification

c) can be feedback or feedforward

d) none of the mentioned

Answer: b

Explanation: It is a case of feedforward networks.

This set of Neural Networks Multiple Choice Questions & Answers  focuses on “Pattern Association – 1″.

1. Feedforward networks are used for?

a) pattern mapping

b) pattern association

c) pattern classification

d) all of the mentioned

Answer: d

Explanation: Feedforward networks are used for pattern mapping, pattern association, pattern classification.

2. Feedback networks are used for?

a) autoassociation

b) pattern storage

c) both autoassociation & pattern storage

d) none of the mentioned

Answer: c

Explanation: Feedback networks are used for autoassociation, pattern storage.

3. The simplest combination network is called competitive learning network?

a) yes

b) no

Answer: a

Explanation: The most basic example of of combination of feedforward & feedback network is competitive learning net.

4. Competitive learning net is used for?

a) pattern grouping

b) pattern storage

c) pattern grouping or storage

d) none of the mentioned

Answer: a

Explanation: Competitive learning net is used for pattern grouping.

5. Feedback connection strength are usually ?

a) fixed

b) variable

c) both fixed or variable type

d) none of the mentioned

Answer: a

Explanation: Feedback connection strength are usually fixed & linear to reduce complexity.

6. Feedforward network are used for pattern storage?

a) yes

b) no

Answer: b

Explanation: Feedforward network are used for pattern mapping, pattern association, pattern classification.

7. If some of output patterns in pattern association problem are identical then problem shifts to?

a) pattern storage problem

b) pattern classification problem

c) pattern mapping problem

d) none of the mentioned

Answer: b

Explanation: Because then number of distinct output can be viewed as class labels.

8. The network for pattern mapping is expected to perform?

a) pattern storage

b) pattern classification

c) genaralization

d) none of the mentioned

Answer: c

Explanation: The network for pattern mapping is expected to perform genaralization.

9. In case of autoassociation by feedback nets in pattern recognition task, what is the behaviour expected?

a) accretive

b) interpolative

c) can be either accretive or interpolative

d) none of the mentioned

Answer: b

Explanation: When a noisy pattern is given , network retrieves a noisy pattern.

10. In case of pattern by feedback nets in pattern recognition task, what is the behaviour expected?

a) accretive

b) interpolative

c) can be either accretive or interpolative

d) none of the mentioned

Answer: a

Explanation:Accretive behaviour is exhibited in case of pattern storage problem.

This set of Neural Networks question bank focuses on “Pattern Association – 2”.

1. What are hard problems?

a) classification problems which are not clearly separable

a) classification problems which are not associatively separable

a) classification problems which are not functionally separable

d) none of the mentioned

Answer: d

Explanation: Classification problems which are not linearly separable separable are known as hard problems.

2. In order to overcome constraint of linearly separablity concept of multilayer feedforward net is proposed?

a) yes

b) no

Answer: a

Explanation: Multilayer feedforward net with non linear processing units in intermidiate hidden layer is proposed.

3. The hard learning problem is ultimately solved by hoff’s algorithm?

a) yes

b) no

Answer: b

Explanation: The hard learning problem is ultimately solved by backpropagation algorithm.

4. What is generalization?

a) ability to store a pattern

b) ability to recall a pattern

c) ability to learn a mapping function

d) none of the mentioned

Answer: c

Explanation: Generalization is the ability to learn a mapping function.

5. Generalization feature of a multilayer feedforward network depends on factors?

a) architectural details

b) learning rate parameter

c) training samples

d) all of the mentioned

Answer: a

Explanation: Generalization feature of a multilayer feedforward network depends on all of these above mentioned factors.

6. What is accretive behaviour?

a) not a type of pattern clustering task

b) for small noise variations pattern lying closet to the desired pattern is recalled.

c) for small noise variations noisy pattern having parameter adjusted according to noise variation is recalled

d) none of the mentioned

Answer: b

Explanation: In accretive behaviour, pattern lying closet to the desired pattern is recalled.

7. What is Interpolative behaviour?

a) not a type of pattern clustering task

b) for small noise variations pattern lying closet to the desired pattern is recalled.

c) for small noise variations noisy pattern having parameter adjusted according to noise variation is recalled

d) none of the mentioned

Answer: c

Explanation: In interpolative behaviour, pattern having parameter adjusted according to noise variation is recalled & not the ideal one.

8. Does pattern association involves non linear units in feedforward neural network?

a) yes

b) no

Answer: b

Explanation: There are only two layers & single set of weights in pattern association.

9. What is the feature that doesn’t belongs to pattern classification in feeddorward neural networks?

a) recall is direct

b) delta rule learning

c) non linear processing units

d) two layers

Answer: b

Explanation: It involves perceptron learning.

10. What is the feature that doesn’t belongs to pattern mapping in feeddorward neural networks?

a) recall is direct

b) delta rule learning

c) non linear processing units

d) two layers

Answer: d

Explanation: It involves multiple layers.

This set of Neural Networks Multiple Choice Questions & Answers  focuses on “Determination of Weights″.

1. In determination of weights by learning, for orthogonal input vectors what kind of learning should be employed?

a) hebb learning law

b) widrow learning law

c) hoff learning law

d) no learning law

Answer: a

Explanation: For orthogonal input vectors, Hebb learning law is best suited.

2. In determination of weights by learning, for linear input vectors what kind of learning should be employed?

a) hebb learning law

b) widrow learning law

c) hoff learning law

d) no learning law

Answer: b

Explanation: For linear input vectors, widrow learning law is best suited.

3. In determination of weights by learning, for noisy input vectors what kind of learning should be employed?

a) hebb learning law

b) widrow learning law

c) hoff learning law

d) no learning law

Answer: d

Explanation: For noisy input vectors, there is no learning law.

4. What are the features that can be accomplished using affine transformations?

a) arbitrary rotation

b) scaling

c) translation

d) all of the mentioned

Answer: d

Explanation: Affine transformations can be used to do arbitrary rotation, scaling, translation.

5. What is the features that cannot be accomplished earlier without affine transformations?

a) arbitrary rotation

b) scaling

c) translation

d) all of the mentioned

Answer: c

Explanation: Affine transformations can be used to do arbitrary rotation, scaling, translation.

6. what are affine transformations?

a) addition of bias term  which results in arbitrary rotation, scaling, translation of input pattern.

b) addition of bias term  which results in arbitrary rotation, scaling, translation of input pattern.

c) addition of bias term  or  which results in arbitrary rotation, scaling, translation of input pattern.

b) none of the mentioned

Answer: a

Explanation: It follows from basic definition of affine transformation.

7. Can a artificial neural network capture association if input patterns is greater then dimensionality of input vectors?

a) yes

b) no

Answer: a

Explanation: By using nonlinear processing units in output layer.

8. By using only linear processing units in output layer, can a artificial neural network capture association if input patterns is greater then dimensionality of input vectors?

a) yes

b) no

Answer: b

Explanation: There is need of non linear processing units.

9. Number of output cases depends on what factor?

a) number of inputs

b) number of distinct classes

c) total number of classes

d) none of the mentioned

Answer: b

Explanation: Number of output cases depends on number of distinct classes.

10. For noisy input vectors, Hebb methodology of learning can be employed?

a) yes

b) no

Answer: b

Explanation: For noisy input vectors, no specific type of learning method exist.

This set of Neural Networks Multiple Choice Questions & Answers  focuses on “Pattern Classification – 1″.

1. What is the objective of perceptron learning?

a) class identification

b) weight adjustment

c) adjust weight along with class identification

d) none of the mentioned

Answer: c

Explanation: The objective of perceptron learning is to adjust weight along with class identification.

2. On what factor the number of outputs depends?

a) distinct inputs

b) distinct classes

c) both on distinct classes & inputs

d) none of the mentioned

Answer: b

Explanation: Number of outputs depends on number of classes.

3. In perceptron learning, what happens when input vector is correctly classified?

a) small adjustments in weight is done

b) large adjustments in weight is done

c) no adjustments in weight is done

d) weight adjustments doesn’t depend on classification of input vector

Answer: c

Explanation: No adjustments in weight is done, since input has been correctly classified which is the objective of the system.

4. When two classes can be separated by a separate line, they are known as?

a) linearly separable

b) linearly inseparable classes

c) may be separable or inseparable, it depends on system

d) none of the mentioned

Answer: a

Explanation: Linearly separable classes, functions can be separated by a line.

5. If two classes are linearly inseparable, can perceptron convergence theorem be applied?

a) yes

b) no

Answer: b

Explanation: Perceptron convergence theorem can only be applied, if and only if two classses are linearly separable.

6. Two classes are said to be inseparable when?

a) there may exist straight lines that doesn’t touch each other

b) there may exist straight lines that can touch each other

c) there is only one straight line that separates them

d) all of the mentioned

Answer: c

Explanation: Linearly separable classes, functions can be separated by a line.

7. Is it necessary to set initial weights in prceptron convergence theorem to zero?

a) yes

b) no

Answer: b

Explanation: Initial setting of weights doesn’t affect perceptron convergence theorem.

8. The perceptron convergence theorem is applicable for what kind of data?

a) binary

b) bipolar

c) both binary and bipolar

d) none of the mentioned

Answer: c

Explanation: The perceptron convergence theorem is applicable for both binary and bipolar input, output data.

9. w = w + n – s) a, where b is desired output, s is actual output, a is input vector and ‘w’ denotes weight, can this model be used for perceptron learning?

a) yes

b) no

Answer: a

Explanation: Gradient descent can be used as perceptron learning.

10. If e denotes error for correction of weight then what is formula for error in perceptron learning model: w = w + n – s) a, where b is desired output, s is actual output, a is input vector and ‘w’ denotes weight

a) e = n – s) a

b) e = n – s)

c) e =  – s)

d) none of the mentioned

Answer: c

Explanation: Error is difference between desired and actual output.

This set of Neural Networks Test focuses on “Pattern Classification – 2”.

1. Convergence in perceptron learning takes place if and only if:

a) a minimal error condition is satisfied

b) actual output is close to desired output

c) classes are linearly separable

d) all of the mentioned

Answer: c

Explanation: Linear separability of classes is the condition for convergence of weighs in perceprton learning.

2. When line joining any two points in the set lies entirely in region enclosed by the set in M-dimensional space , then the set is known as?

a) convex set

b) concave set

c) may be concave or convex

d) none of the mentioned

Answer: a

Explanation: A convex set is a set of points in M-dimensional space such that line joining any two points in the set lies entirely in region enclosed by the set.

3. Is it true that percentage of linearly separable functions will increase rapidly as dimension of input pattern space is increased?

a) yes

b) no

Answer: b

Explanation: There is decrease in number of linearly separable functions as dimension of input pattern space is increased.

4. If pattern classes are linearly separable then hypersurfaces reduces to straight lines?

a) yes

b) no

Answer: a

Explanation: Hypersurfaces reduces to straight lines, if pattern classes are linearly separable.

5. As dimensionality of input vector increases, what happens to linear separability?

a) increases

b) decreases

c) no effect

d) doesn’t depend on dimensionality

Answer: b

Explanation: Linear separability decreases as dimensionality increases.

6. In a three layer network, shape of dividing surface is determined by?

a) number of units in second layer

b) number of units in third layer

c) number of units in second and third layer

d) none of the mentioned

Answer: a

Explanation: Practically, number of units in second layer determines shape of dividing surface.

7. In a three layer network, number of classes is determined by?

a) number of units in second layer

b) number of units in third layer

c) number of units in second and third layer

d) none of the mentioned

Answer: b

Explanation: Practically, number of units in third layer determines number of classes.

8. Intersection of linear hyperplanes in three layer network can only produce convex surfaces, is the statement true?

a) yes

b) no

Answer: a

Explanation: Intersection of linear hyperplanes in three layer network can only produce convex surfaces.

9. Intersection of convex regions in three layer network can only produce convex surfaces, is the statement true?

a) yes

b) no

Answer: b

Explanation: Intersection of convex regions in three layer network can produce nonconvex regions.

10. If the output produces nonconvex regions, then how many layered neural is required at minimum?

a) 2

b) 3

c) 4

d) 5

Answer: c

Explanation: Adding one more layer of units to three layer can yield surfaces which can separate even nonconvex regions.

This set of Neural Networks Multiple Choice Questions & Answers  focuses on “Pattern Mapping″.

1. Can all hard problems be handled by a multilayer feedforward neural network, with nonlinear units?

a) yes

b) no

Answer: a

Explanation: Multilayer perceptrons can deal with all hard problems.

2. What is a mapping problem?

a) when no restrictions such as linear separability is placed on the set of input – output pattern pairs

b) when there may be restrictions such as linear separability placed on input – output patterns

c) when there are restriction but other than linear separability

d) none of the mentioned

Answer: a

Explanation: Its a more general case of classification problem.

3. Can mapping problem be a more general case of pattern classification problem?

a) yes

b) no

Answer: a

Explanation: Since no restrictions such as linear separability is placed on the set of input – output pattern pairs, mapping problem becomes a more general case of pattern classification problem.

4. What is the objective of pattern mapping problem?

a) to capture weights for a link

b) to capture inputs

c) to capture feedbacks

d) to capture implied function

Answer: d

Explanation: The objective of pattern mapping problem is to capture implied function.

5. To provide generalization capability to a network, what should be done?

a) all units should be linear

b) all units should be non – linear

c) except input layer, all units in other layers should be non – linear

d) none of the mentioned

Answer: c

Explanation: To provide generalization capability to a network, except input layer, all units in other layers should be non – linear.

6. What is the objective of pattern mapping problem?

a) to capture implied function

b) to capture system characteristics from observed data

c) both to implied function and system characteristics

d) none of the mentioned

Answer: d

Explanation: The implied fuction is all about system characteristics.

7. Does an approximate system produce strictly an interpolated output?

a) yes

b) no

Answer: b

Explanation: An approximate system doesn’t produce strictly an interpolated output.

8. The nature of mapping problem decides?

a) number of units in second layer

b) number of units in third layer

c) overall number of units in hidden layers

d) none of the mentioned

Answer: c

Explanation: The nature of mapping problem decides overall number of units in hidden layers.

9. How is hard learning problem solved?

a) using nonlinear differentiable output function for output layers

b) using nonlinear differentiable output function for hidden layers

c) using nonlinear differentiable output function for output and hidden layers

d) it cannot be solved

Answer: c

Explanation: Hard learning problem is solved by using nonlinear differentiable output function for output and hidden layers.

10. The number of units in hidden layers depends on?

a) the number of inputs

b) the number of outputs

c) both the number of inputs and outputs

d) the overall characteristics of the mapping problem

Answer: d

Explanation: The number of units in hidden layers depends on the overall characteristics of the mapping problem.

This set of Neural Networks Multiple Choice Questions & Answers  focuses on “Pattern Recognition″.

1. From given input-output pairs pattern recognition model should capture characteristics of the system?

a) true

b) false

Answer: a

Explanation: From given input-output pairs pattern recognition model should be able to capture characteristics of the system & hence should be designed in that manner.

2. Let a, b represent in input-output pairs, where “l” varies in natural range of no.s, then if a=b?

a) problem is heteroassociation

b) problem is autoassociation

c) can be either auto or heteroassociation

d) none of the mentioned

Answer: b

Explanation: When a=b problem is classified as autoassociation.

3. Let a, b represent in input-output pairs, where “l” varies in natural range of no.s, then if a=!b?

a) problem is heteroassociation

b) problem is autoassociation

c) can be either auto or heteroassociation

d) none of the mentioned

Answer: a

Explanation: When a & bare distinct, problem is classified as autoassociation.

4. The recalled output in pattern association problem depends on?

a) nature of input-output

b) design of network

c) both input & design

d) none of the mentioned

Answer: c

Explanation: The recalled output in pattern association problem depends on both input & design of network.

5. If a gives output b & a’=a+m,where m is small quantity & if a’ gives ouput b then?

a) network exhibits accretive behaviour

b) network exhibits interpolative behaviour

c) exhibits both accretive & interpolative behaviour

d) none of the mentioned

Answer: a

Explanation: This follows from basic definition of accretive behaviour in neural.

6. If a gives output b & a’=a+m,where m is small quantity & if a’ gives ouput b+n then?

a) network exhibits accretive behaviour

b) network exhibits interpolative behaviour

c) exhibits both accretive & interpolative behaviour

d) none of the mentioned

Answer: b

Explanation: This follows from basic definition in neural.

7. Can system be both interpolative & accretive at same time?

a) yes

b) no

Answer: b

Explanation: System can’t exhibit both behaviour at same time. since these are based on different approach & algorithm.

8. What are 3 basic types of neural nets that form basic functional units among

i)feedforward ii) loop iii) recurrent iv) feedback v) combination of feed forward & back

a) i, ii, iii

b) i, ii, iv

c) i, iv, v

d) i, iii, v

Answer: c

Explanation: These form the basic functional units of neural nets.

9. Feedback networks are used for autoassociation & pattern storage?

a) yes

b) no

Answer: a

Explanation: Feedback networks are typically used for autoassociation & pattern storage.

10. Feedforward networks are also used for autoassociation & pattern storage?

a) yes

b) no

Answer: b

Explanation: Feedforward networks are used for pattern mapping.

This set of Neural Networks Multiple Choice Questions & Answers  focuses on “Backpropagation Algorithm″.

1. What is the objective of backpropagation algorithm?

a) to develop learning algorithm for multilayer feedforward neural network

b) to develop learning algorithm for single layer feedforward neural network

c) to develop learning algorithm for multilayer feedforward neural network, so that network can be trained to capture the mapping implicitly

d) none of the mentioned

Answer: c

Explanation: The objective of backpropagation algorithm is to to develop learning algorithm for multilayer feedforward neural network, so that network can be trained to capture the mapping implicitly.

2. The backpropagation law is also known as generalized delta rule, is it true?

a) yes

b) no

Answer: a

Explanation: Because it fulfils the basic condition of delta rule.

3. What is true regarding backpropagation rule?

a) it is also called generalized delta rule

b) error in output is propagated backwards only to determine weight updates

c) there is no feedback of signal at nay stage

d) all of the mentioned

Answer: d

Explanation: These all statements defines backpropagation algorithm.

4. There is feedback in final stage of backpropagation algorithm?

a) yes

b) no

Answer: b

Explanation: No feedback is involved at any stage as it is a feedforward neural network.

5. What is true regarding backpropagation rule?

a) it is a feedback neural network

b) actual output is determined by computing the outputs of units for each hidden layer

c) hidden layers output is not all important, they are only meant for supporting input and output layers

d) none of the mentioned

Answer: b

Explanation: In backpropagation rule, actual output is determined by computing the outputs of units for each hidden layer.

6. What is meant by generalized in statement “backpropagation is a generalized delta rule” ?

a) because delta rule can be extended to hidden layer units

b) because delta is applied to only input and output layers, thus making it more simple and generalized

c) it has no significance

d) none of the mentioned

Answer: a

Explanation: The term generalized is used because delta rule could be extended to hidden layer units.

7. What are general limitations of back propagation rule?

a) local minima problem

b) slow convergence

c) scaling

d) all of the mentioned

Answer: d

Explanation: These all are limitations of backpropagation algorithm in general.

8. What are the general tasks that are performed with backpropagation algorithm?

a) pattern mapping

b) function approximation

c) prediction

d) all of the mentioned

Answer: d

Explanation: These all are the tasks that can be performed with backpropagation algorithm in general.

9. Does backpropagaion learning is based on gradient descent along error surface?

a) yes

b) no

c) cannot be said

d) it depends on gradient descent but not error surface

Answer: a

Explanation: Weight adjustment is proportional to negative gradient of error with respect to weight.

10. How can learning process be stopped in backpropagation rule?

a) there is convergence involved

b) no heuristic criteria exist

c) on basis of average gradient value

d) none of the mentioned

Answer: c

Explanation: If average gadient value fall below a preset threshold value, the process may be stopped.

This set of Neural Networks Multiple Choice Questions & Answers  focuses on “Analysis Of Pattern Storage″.

1. Which is a simplest pattern recognition task in a feedback network?

a) heteroassociation

b) autoassociation

c) can be hetero or autoassociation, depends on situation

d) none of the mentioned

Answer: b

Explanation: Autoassociation is the simplest pattern recognition task.

2. In a linear autoassociative network, if input is noisy than output will be noisy?

a) yes

b) no

Answer: a

Explanation: Linear autoassociative network gives out, what is given to it as input.

3. Does linear autoassociative network have any practical use?

a) yes

b) no

Answer: b

Explanation: Since if input is noisy then output will aslo be noisy, hence no practical use.

4. What can be done by using non – linear output function for each processing unit in a feedback network?

a) pattern classification

b) recall

c) pattern storage

d) all of the mentioned

Answer: c

Explanation: By using non – linear output function for each processing unit, a feedback network can be used for pattern storage.

5. When are stable states reached in energy landscapes, that can be used to store input patterns?

a) mean of peaks and valleys

b) maxima

c) minima

d) none of the mentioned

Answer: c

Explanation: Energy minima corresponds to stable states that can be used to store input patterns.

6. The number of patterns that can be stored in a given network depends on?

a) number of units

b) strength of connecting links

c) both number of units and strength of connecting links

d) none of the mentioned

Answer: c

Explanation: The number of patterns that can be stored in a given network depends on number of units and strength of connecting links.

7. What happens when number of available energy minima be less than number of patterns to be stored?

a) pattern storage is not possible in that case

b) pattern storage can be easily done

c) pattern storage problem becomes hard problem for the network

d) none of the mentioned

Answer: c

Explanation: Pattern storage problem becomes hard problem, when number of energy minima i.e stable states are less.

8. What happens when number of available energy minima be more than number of patterns to be stored?

a) no effect

b) pattern storage is not possible in that case

c) error in recall

d) none of the mentioned

Answer: c

Explanation: Due to additional false minima, there is error in recall.

9. How hard problem can be solved?

a) by providing additional units in a feedback network

b) nothing can be done

c) by removing units in hidden layer

d) none of the mentioned

Answer: a

Explanation: Hard problem can be solved by providing additional units in a feedback network.

10. Why there is error in recall, when number of energy minima is more the required number of patterns to be stored?

a) due to noise

b) due to additional false maxima

c) due to additional false minima

d) none of the mentioned

Answer: c

Explanation: Due to additional false minima, there is error in recall.

This set of Neural Networks Multiple Choice Questions & Answers  focuses on “Introduction Of Feedback Neural Network″.

1. How can false minima be reduced in case of error in recall in feedback neural networks?

a) by providing additional units

b) by using probabilistic update

c) can be either probabilistic update or using additional units

d) none of the mentioned

Answer: b

Explanation: Hard problem can be solved by additional units not the false minima.

2. What is a Boltzman machine?

a) A feedback network with hidden units

b) A feedback network with hidden units and probabilistic update

c) A feed forward network with hidden units

d) A feed forward network with hidden units and probabilistic update

Answer: b

Explanation: Boltzman machine is a feedback network with hidden units and probabilistic update.

3. What is objective of linear autoassociative feedforward networks?

a) to associate a given pattern with itself

b) to associate a given pattern with others

c) to associate output with input

d) none of the mentioned

Answer: a

Explanation: The objective of linear autoassociative feedforward networks is to associate a given pattern with itself.

4. Is there any error in linear autoassociative networks?

a) yes

b) no

Answer: b

Explanation: Because input comes out as output.

5. If input is ‘ a + e ‘ where ‘e’ is the noise introduced, then what is the output in case of autoassociative feedback network?

a) a

b) a + e

c) could be either a or a + e

d) e

Answer: b

Explanation: This is due to the absence of accretive behaviour.

6. If input is ‘ a + e ‘ where ‘e’ is the noise introduced, then what is the output if system is accretive in nature?

a) a

b) a + e

c) could be either a or a + e

d) e

Answer: a

Explanation: This is the property of accretive system.

7. If input is ‘ a + e ‘ where ‘e’ is the noise introduced, then what is the output if system is interpolative in nature?

a) a

b) a + e

c) could be either a or a + e

d) e

Answer: b

Explanation: This is the property of interpolative system.

8. What property should a feedback network have, to make it useful for storing information?

a) accretive behaviour

b) interpolative behaviour

c) both accretive and interpolative behaviour

d) none of the mentioned

Answer: a

Explanation: During recall accretive behaviour make it possible for system to store information.

9. What is the objective of a pattern storage task in a network?

a) to store a given set of patterns

b) to recall a give set of patterns

c) both to store and recall

d) none of the mentioned

Answer: c

Explanation: The objective of a pattern storage task in a network is to store and recall a given set of patterns.

10. Linear neurons can be useful for application such as interpolation, is it true?

a) yes

b) no

Answer: a

Explanation: This means for input vector x, output vector y is produced and for input a.x, output will be a.y.

This set of Neural Networks Multiple Choice Questions & Answers  focuses on “ Analysis of Linear Autoassociative FF Network″.

1. What is the objective of pattern recall?

a) it should not take place when relations are disturbed

b) it should take place when relations are slightly disturbed

c) there is no such objective of recall, it depends on the system

d) none of the mentioned

Answer: b

Explanation: The pattern recall should take place even though features and their spatial relations are slightly disturbed due to noise.

2. Can data be stored through weights?

a) yes

b) no

Answer: a

Explanation: Data can be stored through weights as in case of binary patterns.

3. How is pattern storage task generally accomplished?

a) by a feedback network consisting of processing units with non linear output functions

b) by a feedback network consisting of processing units with linear output functions

c) by a feedforward network consisting of processing units with non linear output functions

d) by a feedforward network consisting of processing units with linear output functions

Answer: b

Explanation: Pattern storage task generally accomplished by a feedback network consisting of processing units with non linear output functions.

4. The trajectory of the state is determined by?

a) activation dynamics

b) synaptic dynamics

c) both activation and synaptic dynamics

d) none of the mentioned

Answer: a

Explanation: The trajectory of the state is determined by activation dynamics.

5. what do you mean by the term trajectory of states?

a) just a state of the network

b) sates at energy minima

c) states at energy maxima

d) none of the mentioned

Answer: d

Explanation: The term trajectory of states means state of the network at successive instants of time.

6. What determines shape of energy landscape?

a) network parameters

b) network states

c) both network parameter and states

d) none of the mentioned

Answer: c

Explanation: The shape of energy landscape is determined by network parameter and states.

7. What may create basins of attraction in energy landscape?

a) feedback among units

b) nonlinear processing in units

c) both feedback and nonlinear processing in units

d) none of the mentioned

Answer: c

Explanation: Feedback and nonlinear processing in units may create basins of attraction in energy landscape.

8. What is the effect of basins of attraction on energy landscape?

a) leads to small deviations

b) leads to fluctuation around

c) may lead to deviation or fluctuation depends on external noise

d) none of the mentioned

Answer: a

Explanation: Basins of attraction in energy landscape leads to small deviations.

9. What does basins of attraction corresponds to?

a) stable states

b) unstable states

c) neutral states

d) none of the mentioned

Answer: a

Explanation: Basins of attraction corresponds to the regions of stable equilibrium states.

10. Basins of attraction in energy landscape leads to fluctuations, is that true?

a) yes

b) no

Answer: b

Explanation: Basins of attraction in energy landscape leads to only small deviations.

This set of Neural Networks Multiple Choice Questions & Answers  focuses on “Analysis Of Pattern Storage Networks – 1″.

1. For what purpose energy minima are used?

a) pattern classification

b) patten mapping

c) pattern storage

d) none of the mentioned

Answer: c

Explanation: Energy minima are used for pattern storage.

2. Is it possible to determine exact number of basins of attraction in energy landscape?

a) yes

b) no

Answer: b

Explanation: It is not possible to determine exact number of basins of attraction in energy landscape.

3. What is capacity of a network?

a) number of inputs it can take

b) number of output it can deliver

c) number of patterns that can be stored

d) none of the mentioned

Answer: c

Explanation: The capacity of a network is the number of patterns that can be stored.

4. Can probability of error in recall be reduced?

a) yes

b) no

Answer: a

Explanation: Probability of error in recall be reduced by adjusting weights in such a way that it is matched to probability distribution of desired patterns.

5. Number of desired patterns is what of basins of attraction?

a) dependent

b) independent

c) dependent or independent

d) none of the mentioned

Answer: b

Explanation: Number of desired patterns is independent of basins of attraction.

6. What happens when number of patterns is more than number of basins of attraction?

a) false wells

b) storage problem becomes hard problem

c) no storage and recall can take place

d) none of the mentioned

Answer: b

Explanation: When number of patterns is more than number of basins of attraction then storage problem becomes hard problem.

7. What happens when number of patterns is less than number of basins of attraction?

a) false wells

b) storage problem becomes hard problem

c) no storage and recall can take place

d) none of the mentioned

Answer: a

Explanation: False wells are created when number of patterns is less than number of basins of attraction.

8. What is hopfield model?

a) fully connected feedback network

b) fully connected feedback network with symmetric weights

c) fully connected feedforward network

d) fully connected feedback network with symmetric weights

Answer: b

Explanation: Hopfield model is fully connected feedback network with symmetric weights.

9. When are false wells created?

a) when number of patterns is more than number of basins of attraction

b) when number of patterns is less than number of basins of attraction

c) when number of patterns is same as number of basins of attraction

d) none of the mentioned

Answer: b

Explanation: False wells are created when number of patterns is less than number of basins of attraction.

10. When does storage problem becomes hard problem?

a) when number of patterns is more than number of basins of attraction

b) when number of patterns is less than number of basins of attraction

c) when number of patterns is same as number of basins of attraction

d) none of the mentioned

Answer: a

Explanation: When number of patterns is more than number of basins of attraction then storage problem becomes hard problem.

This set of Neural Networks Quiz focuses on “Analysis Of Pattern Storage Networks – 2”.

1. For what purpose energy minima are used?

a) pattern classification

b) patten mapping

c) pattern storage

d) none of the mentioned

Answer: c

Explanation: Energy minima are used for pattern storage.

2. Is it possible to determine exact number of basins of attraction in energy landscape?

a) yes

b) no

Answer: b

Explanation: It is not possible to determine exact number of basins of attraction in energy landscape.

3. What is capacity of a network?

a) number of inputs it can take

b) number of output it can deliver

c) number of patterns that can be stored

d) none of the mentioned

Answer: c

Explanation: The capacity of a network is the number of patterns that can be stored.

4. Can probability of error in recall be reduced?

a) yes

b) no

Answer: a

Explanation: Probability of error in recall be reduced by adjusting weights in such a way that it is matched to probability distribution of desired patterns.

5. Number of desired patterns is what of basins of attraction?

a) dependent

b) independent

c) dependent or independent

d) none of the mentioned

Answer: b

Explanation: Number of desired patterns is independent of basins of attraction.

6. What happens when number of patterns is more than number of basins of attraction?

a) false wells

b) storage problem becomes hard problem

c) no storage and recall can take place

d) none of the mentioned

Answer: b

Explanation: When number of patterns is more than number of basins of attraction then storage problem becomes hard problem.

7. What happens when number of patterns is less than number of basins of attraction?

a) false wells

b) storage problem becomes hard problem

c) no storage and recall can take place

d) none of the mentioned

Answer: a

Explanation: False wells are created when number of patterns is less than number of basins of attraction.

8. What is hopfield model?

a) fully connected feedback network

b) fully connected feedback network with symmetric weights

c) fully connected feedforward network

d) fully connected feedback network with symmetric weights

Answer: b

Explanation: Hopfield model is fully connected feedback network with symmetric weights.

9. When are false wells created?

a) when number of patterns is more than number of basins of attraction

b) when number of patterns is less than number of basins of attraction

c) when number of patterns is same as number of basins of attraction

d) none of the mentioned

Answer: b

Explanation: False wells are created when number of patterns is less than number of basins of attraction.

10. When does storage problem becomes hard problem?

a) when number of patterns is more than number of basins of attraction

b) when number of patterns is less than number of basins of attraction

c) when number of patterns is same as number of basins of attraction

d) none of the mentioned

Answer: a

Explanation: When number of patterns is more than number of basins of attraction then storage problem becomes hard problem.

This set of Neural Networks Multiple Choice Questions & Answers  focuses on “Hopfield Model – 1″.

1. How can states of units be updated in hopfield model?

a) synchronously

b) asynchronously

c) synchronously and asynchronously

d) none of the mentioned

Answer: c

Explanation: States of units be updated synchronously and asynchronously in hopfield model.

2. What is synchronous update in hopfield model?

a) all units are updated simultaneously

b) a unit is selected at random and its new state is computed

c) a predefined unit is selected and its new state is computed

d) none of the mentioned

Answer: a

Explanation: In synchronous update, all units are updated simultaneously.

3. What is asynchronous update in hopfield model?

a) all units are updated simultaneously

b) a unit is selected at random and its new state is computed

c) a predefined unit is selected and its new state is computed

d) none of the mentioned

Answer: b

Explanation: In asynchronous update, a unit is selected at random and its new state is computed.

4. Asynchronous update ensures that the next state is atmost unit hamming distance from current state, is that true?

a) yes

b) no

Answer: a

Explanation: Asynchronous update ensures that the next state is at most unit hamming distance from current state.

5. If pattern is to be stored, then what does stable state should have updated value of?

a) current sate

b) next state

c) both current and next state

d) none of the mentioned

Answer: a

Explanation: Stable state should have updated value of current sate.

6. For symmetric weights there exist?

a) basins of attraction corresponding to energy minimum

b) false wells

c) fluctuations in energy landscape

d) none of he mentioned

Answer: a

Explanation: For symmetric weights there exist a stable point.

7. If connections are not symmetric then basins of attraction may correspond to?

a) oscillatory regions

b) stable regions

c) chaotic regions

d) oscillatory or chaotic regions

Answer: d

Explanation: If connections are not symmetric then basins of attraction may correspond to oscillatory or chaotic regions.

8. For analysis of storage capacity what are the conditions imposed on hopfield model?

a) symmetry of weights

b) asynchronous update

c) symmetry of weights and asynchronous update

d) none of the mentioned

Answer: c

Explanation: For analysis of storage capacity, symmetry of weights and asynchronous update conditions are imposed on hopfield model.

9. What is gradient descent?

a) method to find the absolute minimum of a function

b) method to find the absolute maximum of a function

c) maximum or minimum, depends on the situation

d) none of the mentioned

Answer: a

Explanation: Gradient descent gives absolute minimum of a function.

10. If connections are not symmetric then basins of attraction may correspond to oscillatory or stable regions, is that true?

a) yes

b) no

Answer: b

Explanation: Asymmetric weight can’t lead to stable regions.

This set of Neural Networks MCQs focuses on “Hopfield Model – 2”.

1. In hopfield network with symmetric weights, energy at each state may?

a) increase

b) decrease

c) decrease or remain same

d) decrease or increase

Answer: c

Explanation: Energy of the network cant increase as it may then lead to instability.

2. In hopfield model with symmetric weights, network can move to?

a) lower

b) higher

c) lower or higher

d) lower or same

Answer: d

Explanation: In hopfield model with symmetric weights, network can move to lower or same state.

3. Can error in recall due to false minima be reduced?

a) yes

b) no

Answer: a

Explanation: There are generally two methods to reduce error in recall due to false minima.

4. How can error in recall due to false minima be reduced?

a) deterministic update for states

b) stochastic update for states

c) not possible

d) none of the mentioned

Answer: b

Explanation: Error in recall due to false minima can be reduced by stochastic update for states.

5. Energy at each state in hopfield with symmetric weights network may increase or decrease?

a) yes

b) no

Answer: b

Explanation: Energy of the network cant increase as it may then lead to instability.

6. Pattern storage problem which cannot be represented by a feedback network of given size can be called as?

a) easy problems

b) hard problems

c) no such problem exist

d) none of the mentioned

Answer: b

Explanation: Pattern storage problem which cannot be represented by a feedback network of given size are known as hard problems.

7. What is the other way to reduce error in recall due to false minima apart from stochastic update?

a) no other method exist

b) by storing desired patterns at lowest energy minima

c) by storing desired patterns at energy maxima

d) none of the mentioned

Answer: b

Explanation: Error in recall due to false minima can be reduced by stochastic update or by storing desired patterns at lowest energy minima.

8. How can error in recall due to false minima be further reduced?

a) using suitable activation dynamics

b) cannot be further reduced

c) by storing desired patterns at energy maxima

d) none of the mentioned

Answer: a

Explanation: Error in recall due to false minima can further be reduced by using suitable activation dynamics.

9. As temperature increase, what happens to stochastic update?

a) increase in update

b) decrease in update

c) no change

d) none of the mentioned

Answer: c

Explanation: Temperature doesn’t effect stochastic update.

10. Why does change in temperature doesn’t effect stochastic update?

a) shape landscape depends on the network and its weights which varies accordingly and compensates the effect

b) shape landscape depends on the network and its weights which is fixed

c) shape landscape depends on the network, its weights and the output function which varies accordingly and compensates the effect

d) shape landscape depends on the network, its weights and the output function which is fixed

Answer: d

Explanation: Change in temperature doesn’t effect stochastic update because shape landscape depends on the network, its weights and the output function which is fixed.

This set of Neural Networks Multiple Choice Questions & Answers  focuses on “Stochastic Networks″.

1. p = 1/)) ,where ‘s’ is the output given the activation ‘x’ is a?

a) hopfield network

b) sigma network

c) stochastic network

d) none of the mentioned

Answer: c

Explanation: This is the basic equation of a stochastic network.

2. Does a stochastic network will evolve differently each time it is run?

a) yes

b) no

Answer: a

Explanation: As trajectory of the state of the network becomes a sample function of a random process.

3. In case of deterministic update, what kind of equilibrium is reached?

a) static

b) dynamic

c) neutral

d) none of the mentioned

Answer: a

Explanation: In case of deterministic update, static equilibrium is reached.

4. In case of stochastic update, can static equilibrium be reached?

a) yes

b) no

Answer: b

Explanation: There will never be a static equilibrium in stochastic network.

5. In case of stochastic update, what kind of equilibrium is reached?

a) static

b) dynamic

c) neutral

d) equilibrium not possible

Answer: b

Explanation: In case of stochastic update, dynamic equilibrium is reached.

6. Is it possible in stochastic network that average state of network doesn’t change with time?

a) yes

b) no

Answer: a

Explanation: Dynamic equilibrium is possible in stochastic network.

7. What can be the possible reason for thermal equilibrium in stochastic networks?

a) probability distribution of states changes and compensates

b) probability distribution change with only update

c) probability distribution does not change with time

d) none of the mentionedstochastic network exhibits stable states

Answer: c

Explanation: Probability distribution does not change with time is the only reason for thermal equilibrium in stochastic networks.

8. Can networks with symmetric weight reach thermal equilibrium?

a) yes

b) no

Answer: a

Explanation: Networks with symmetric weight reach thermal equilibrium at a given temperature.

9. When activation value is determined by using the average of fluctuations of outputs from other units, it is known as?

a) maximum field approximation

b) median field approximation

c) minimum field approximation

d) none of the mentioned

Answer: d

Explanation: It is known as mean field approximation.

10. Where does a stochastic network exhibits stable states ?

a) at any temperature

b) above critical temperature

c) at critical temperature

d) below critical temperature

Answer: d

Explanation: Stochastic network exhibits stable states below critical temperature.

This set of Neural Networks Multiple Choice Questions & Answers  focuses on “Boltzman Machine – 1″.

1. Probability of error in recall of stored patterns can be reduced if?

a) patterns are stored appropriately

b) inputs are captured appropriately

c) weights are chosen appropriately

d) none of the mentioned

Answer: c

Explanation: Probability of error in recall of stored patterns can be reduced if weights are chosen appropriately.

2. What is pattern environment?

a) probability of desired patterns

b) probability of given patterns

c) behaviour of system

d) none of the mentioned

Answer: d

Explanation: Pattern environment is probability distribution of given patterns.

3. For what purpose is pattern environment useful?

a) determining structure

b) determining desired outputs

c) determining future inputs

d) none of the mentioned

Answer: d

Explanation: Pattern environment is useful for determining weights.

4. What should be the aim of training procedure in boltzman machine of feedback networks?

a) to capture inputs

b) to feedback the captured outputs

c) to capture the behaviour of system

d) none of the mentioned

Answer: d

Explanation: The training procedure should try to capture the pattern environment.

5. What consist of boltzman machine?

a) fully connected network with both hidden and visible units

b) asynchronous operation

c) stochastic update

d) all of the mentioned

Answer: d

Explanation: Boltzman machine consist of fully connected network with both hidden and visible units operating asynchronously with stochastic update.

6. By using which method, boltzman machine reduces effect of additional stable states?

a) no such method exist

b) simulated annealing

c) hopfield reduction

d) none of the mentioned

Answer: b

Explanation: boltzman machine uses simulated annealing to reduce the effect of additional stable states.

7. For which other task can boltzman machine be used?

a) pattern mapping

b) feature mapping

c) classification

d) pattern association

Answer: d

Explanation: Boltzman machine can be used for pattern association.

8. How are energy minima related to probability of occurrence of corresponding patterns in the environment?

a) directly

b) inversely

c) directly or inversely

d) no relation

Answer: a

Explanation: Energy minima is directly related to probability of occurrence of corresponding patterns in the environment.

9. Is exact representation of pattern environment possible?

a) yes

b) no

Answer: b

Explanation: Exact representation of pattern environment is not possible.

10. What may be the reasons for non zero probability of error in recalling?

a) spurious stable states

b) approximation in pattern environment representation

c) extra stable states

d) all of the mentioned

Answer: d

Explanation: These all are the primary reasons for existence of non zero probability of error.

This set of Neural Networks Multiple Choice Questions & Answers focuses on “Boltzman Machine – 2”.

1. For what purpose Feedback neural networks are primarily used?

a) classification

b) feature mapping

c) pattern mapping

d) none of the mentioned

Answer: d

Explanation: Feedback neural networks are primarily used for pattern storage.

2. Presence of false minima will have what effect on probability of error in recall?

a) directly

b) inversely

c) no effect

d) directly or inversely

Answer: a

Explanation: Presence of false minima will increase the probability of error in recall.

3. How is effect false minima reduced

a) deterministic update of weights

b) stochastic update of weights

c) deterministic or stochastic update of weights

d) none of the mentioned

Answer: b

Explanation: Presence of false minima can be reduced by stochastic update.

4. Is Boltzman law practical for implementation?

a) yes

b) no

Answer: b

Explanation: Boltzman law is too slow for implementation.

5. For practical implementation what type of approximation is used on boltzman law?

a) max field approximation

b) min field approximation

c) hopfield approximation

d) none of the mentioned

Answer: d

Explanation: For practical implementation mean field approximation is used.

6. What happens when we use mean field approximation with boltzman learning?

a) it slows down

b) it get speeded up

c) nothing happens

d) may speedup or speed down

Answer: b

Explanation: Boltzman learning get speeded up using mean field approximation.

7. Approximately how much times the boltzman learning get speeded up using mean field approximation?

a) 5-10

b) 10-30

c) 30-50

d) 50-70

Answer: b

Explanation: Boltzman learning get speeded up 10-30 using mean field approximation.

8.False minima can be reduced by deterministic updates?

a) yes

b) no

Answer: b

Explanation: Presence of false minima can be reduced by stochastic update.

9. In boltzman learning which algorithm can be used to arrive at equilibrium?

a) hopfield

b) mean field

c) hebb

d) none of the mentioned

Answer: d

Explanation: Metropolis algorithm can be used to arrive at equilibrium.

10. Boltzman learning is a?

a) fast process

b) steady process

c) slow process

d) none of the mentioned

Answer: d

Explanation: Boltzman learning is a slow process.

This set of Neural Networks Multiple Choice Questions & Answers  focuses on “Competitive Learning Neural Nework Introduction″.

1. How are input layer units connected to second layer in competitive learning networks?

a) feedforward manner

b) feedback manner

c) feedforward and feedback

d) feedforward or feedback

Answer: a

Explanation: The output of input layer is given to second layer with adaptive feedforward weights.

2. Which layer has feedback weights in competitive neural networks?

a) input layer

b) second layer

c) both input and second layer

d) none of the mentioned

Answer: b

Explanation: Second layer has weights which gives feedback to the layer itself.

3. What is the nature of general feedback given in competitive neural networks?

a) self excitatory

b) self inhibitory

c) self excitatory or self inhibitory

d) none of the mentioned

Answer: a

Explanation: The output of each unit in second layer is fed back to itself in self – excitatory manner.

4. What consist of competitive learning neural networks?

a) feedforward paths

b) feedback paths

c) either feedforward or feedback

d) combination of feedforward and feedback

Answer: Competitive learning neural networks is a combination of feedforward and feedback connection layers resulting in some kind of competition.

5. What conditions are must for competitive network to perform pattern clustering?

a) non linear output layers

b) connection to neighbours is excitatory and to the farther units inhibitory

c) on centre off surround connections

d) none of the mentioned fulfils the whole criteria

Answer: d

Explanation: If the output functions of units in feedback laye are made non-linear , with fixed weight on-centre off-surround connections, the pattern clustering can be performed.

6. What conditions are must for competitive network to perform feature mapping?

a) non linear output layers

b) connection to neighbours is excitatory and to the farther units inhibitory

c) on centre off surround connections

d) none of the mentioned fulfils the whole criteria

Answer: d

Explanation: If cndition in a, b, c are met then feature mapping can be performed.

7. If a competitive network can perform feature mapping then what is that network can be called?

a) self excitatory

b) self inhibitory

c) self organization

d) none of the mentioned

Answer: c

Explanation: Competitive network that can perform feature mapping can be called as self organization network.

8. What is an instar?

a) receives inputs from all others

b) gives output to all others

c) may receive or give input or output to others

d) none of the mentioned

Answer: a

Explanation: An instar receives inputs from all other input units.

9. How is weight vector adjusted in basic competitive learning?

a) such that it moves towards the input vector

b) such that it moves away from input vector

a) such that it moves towards the output vector

b) such that it moves away from output vector

Answer: a

Explanation: Weight vector is adjusted such that it moves towards the input vector.

Answer: a

Explanation: The update in weight vector in basic competitive learning can be represented by w = w + del.w.

This set of Neural Networks Multiple Choice Questions & Answers  focuses on “Feedback Layer″.

1. An instar can respond to a set of input vectors even if its not trained to capture the behaviour of the set?

a) yes

b) no

Answer: a

Explanation: An instar can respond to a set of input vectors even if it is trained to capture the average behaviour of the set.

2. The weight change in plain hebbian learning is?

a) 0

b) 1

c) 0 or 1

d) none of the mentioned

Answer: d

Explanation: The weight change in plain hebbian learning can never be zero.

3. What is the nature of weights in plain hebbian learning?

a) convergent

b) divergent

c) may be convergent or divergent

d) none of the mentioned

Answer: b

Explanation: In plain hebbian learning weights keep growing without bound.

4. How can divergence be prevented?

a) using hopfield criteria

b) sangers rule

c) ojas rule

d) sangers or ojas rule

Answer: d

Explanation: Divergence can be prevented by using sangers or ojas rule.

5. By normalizing the weight at every stage can we prevent divergence?

a) yes

b) no

Answer: a

Explanation: ||w|| = 1 .

6. What is ojas rule?

a) finds a unit weight vector

b) maximises the mean squared output

c) minimises the mean squared output

d) none of the mentioned

Answer: d

Explanation: Ojas rule finds a unit weight vector and maximises the mean squared output.

7. What is the other name of feedback layer in competitive neural networks?

a) feedback layer

b) feed layer

c) competitive layer

d) no such name exist

Answer: c

Explanation: Feedback layer in competitive neural networks is also known as competitive layer.

8. what kind of feedbacks are given in competitive layer?

a) self excitatory to self and others

b) inhibitory to self and others

c) self excitatory to self and inhibitory to others

d) inhibitory to self and excitatory to others

Answer: c

Explanation: The second layer of competitive networks have self excitatory to self and inhibitory to others feedbacks to make it competitive.

9. Generally how many kinds of pattern storage network exist?

a) 2

b) 3

c) 4

d) 5

Answer: b

Explanation: Namely, temporary storage, Short term memory, Long term memory.

10. In competitive learning, node with highest activation is the winner, is it true?

a) yes

b) no

Answer: a

Explanation: This itself defines the competitive learning.

This set of Neural Networks Multiple Choice Questions & Answers  focuses on “Analysis of Feature Mapping Network″.

1. What kind of learning is involved in pattern clustering task?

a) supervised

b) unsupervised

c) learning with critic

d) none of the mentioned

Answer: b

Explanation: Since pattern classes are formed on unlabelled classes.

2. In pattern clustering, does physical location of a unit relative to other unit has any significance?

a) yes

b) no

c) depends on type of clustering

d) none of the mentioned

Answer: b

Explanation: Physical location of a unit doesn’t effect the output.

3. How is feature mapping network distinct from competitive learning network?

a) geometrical arrangement

b) significance attached to neighbouring units

c) nonlinear units

d) none of the mentioned

Answer: d

Explanation: Both the geometrical arrangement and significance attached to neighbouring units make it distinct.

4. What is the objective of feature maps?

a) to capture the features in space of input patterns

b) to capture just the input patterns

c) update weights

d) to capture output patterns

Answer: a

Explanation: The objective of feature maps is to capture the features in space of input patterns.

5. How are weights updated in feature maps?

a) updated for winning unit only

b) updated for neighbours of winner only

c) updated for winning unit and its neighbours

d) none of the mentioned

Answer: c

Explanation: Weights are updated in feature maps for winning unit and its neighbours.

6. In feature maps, when weights are updated for winning unit and its neighbour, which type learning it is known as?

a) karnaugt learning

b) boltzman learning

c) kohonen’s learning

d) none of the mentioned

Answer: c

Explanation: Self organization network is also known as Kohonen learning.

7. In self organizing network, how is layer connected to output layer?

a) some are connected

b) all are one to one connected

c) each input unit is connected to each output unit

d) none of the mentioned

Answer: c

Explanation: In self organizing network, each input unit is connected to each output unit.

8. What is true regarding adaline learning algorithm

a) uses gradient descent to determine the weight vector that leads to minimal error

b) error is defined as MSE between neurons net input and its desired output

c) this technique allows incremental learning

d) all of the mentioned

Answer: d

Explanation: Incremental learning means refining of the weights as more training samples are added, rest are basic statements that defines adaline learning.

9. What is true for competitive learning?

a) nodes compete for inputs

b) process leads to most efficient neural representation of input space

c) typical for unsupervised learning

d) all of the mentioned

Answer: d

Explanation: These all statements defines the competitive learning.

10. Use of nonlinear units in the feedback layer of competitive network leads to concept of?

a) feature mapping

b) pattern storage

c) pattern classification

d) none of the mentioned

Answer: d

Explanation: Use of nonlinear units in the feedback layer of competitive network leads to concept of pattern clustering.

This set of Neural Networks Multiple Choice Questions & Answers  focuses on “Associative Memories″.

1. What are the tasks that cannot be realised or recognised by simple networks?

a) handwritten characters

b) speech sequences

c) image sequences

d) all of the mentioned

Answer: d

Explanation: These all are complex recognition tasks.

2. Can data be stored directly in associative memory?

a) yes

b) no

Answer: b

Explanation: Data cannot be stored directly in associative memory.

3. If the weight matrix stores the given patterns, then the network becomes?

a) autoassoiative memory

b) heteroassociative memory

c) multidirectional assocative memory

d) temporal associative memory

Answer: a

Explanation: If the weight matrix stores the given patterns, then the network becomes autoassoiative memory.

4. If the weight matrix stores association between a pair of patterns, then network becomes?

a) autoassoiative memory

b) heteroassociative memory

c) multidirectional assocative memory

d) temporal associative memory

Answer: b

Explanation: If the weight matrix stores the given patterns, then the network becomes heteroassociative memory.

5. If the weight matrix stores multiple associations among several patterns, then network becomes?

a) autoassoiative memory

b) heteroassociative memory

c) multidirectional assocative memory

d) temporal associative memory

Answer: a

Explanation: If the weight matrix stores the given patterns, then the network becomes multidirectional assocative memory.

6. If the weight matrix stores association between adjacent pairs of patterns, then network becomes?

a) autoassoiative memory

b) heteroassociative memory

c) multidirectional assocative memory

d) temporal associative memory

Answer: a

Explanation: If the weight matrix stores the given patterns, then the network becomes temporal associative memory.

7. Heteroassociative memory is also known as?

a) unidirectional memory

b) bidirectional memory

c) multidirectional assocative memory

d) temporal associative memory

Answer: b

Explanation: Heteroassociative memory is also known as bidirectional memory.

8. What are some of desirable characteristics of associative memories?

a) ability to store large number of patterns

b) fault tolerance

c) able to recall, even for input pattern is noisy

d) all of the mentioned

Answer: d

Explanation: These all are desirable characteristics of associative memories.

9. What is the objective of BAM?

a) to store pattern pairs

b) to recall pattern pairs

c) to store a set of pattern pairs and they can be recalled by giving either of pattern as input

d) none of the mentioned

Answer: c

Explanation: The objective of BAM i.e Bidirectional Associative Memory, is to store a set of pattern pairs and they can be recalled by giving either of pattern as input.

10. BAM is a special case of MAM, is that true?

a) yes

b) no

Answer: a

Explanation: BAM i.e Bidirectional Associative Memory is a special case of MAM i.e Multidirectional Associative Memory.

This set of Neural Networks Multiple Choice Questions & Answers  focuses on “Multi Layer Feedforward Neural Network″.

1. What is the use of MLFFNN?

a) to realize structure of MLP

b) to solve pattern classification problem

c) to solve pattern mapping problem

d) to realize an approximation to a MLP

Answer: d

Explanation: MLFFNN stands for multilayer feedforward network and MLP stands for multilayer perceptron.

2. What is the advantage of basis function over mutilayer feedforward neural networks?

a) training of basis function is faster than MLFFNN

b) training of basis function is slower than MLFFNN

c) storing in basis function is faster than MLFFNN

d) none of the mentioned

Answer: a

Explanation: The main advantage of basis function is that the training of basis function is faster than MLFFNN.

3. Why is the training of basis function is faster than MLFFNN?

a) because they are developed specifically for pattern approximation

b) because they are developed specifically for pattern classification

c) because they are developed specifically for pattern approximation or classification

d) none of the mentioned

Answer: c

Explanation: Training of basis function is faster than MLFFNN because they are developed specifically for pattern approximation or classification.

4. Pattern recall takes more time for?

a) MLFNN

b) Basis function

c) Equal for both MLFNN and basis function

d) None of the mentioned

Answer: b

Explanation: The first layer of basis function involves computations.

5. In which type of networks training is completely avoided?

a) GRNN

b) PNN

c) GRNN and PNN

d) None of the mentioned

Answer: c

Explanation: In GRNN and PNN networks training is completely avoided.

6. What does GRNN do?

a) function approximation task

b) pattern classification task

c) function approximation and pattern classification task

d) none of the mentioned

Answer: a

Explanation: GRNN stand for Generalized Regression Neural Networks.

7. What does PNN do?

a) function approximation task

b) pattern classification task

c) function approximation and pattern classification task

d) none of the mentioned

Answer: b

Explanation: PNN stand for Probabilistic Neural Networks.

8. Th CPN provides practical approach for implementing?

a) patter approximation

b) pattern classification

c) pattern mapping

d) pattern clustering

Answer: c

Explanation: CPN i.e counterpropagation network provides a practical approach for implementing pattern mapping.

9. What consist of a basic counterpropagation network?

a) a feedforward network only

b) a feedforward network with hidden layer

c) two feedforward network with hidden layer

d) none of the mentioned

Answer: c

Explanation: Counterpropagation network consist of two feedforward network with a common hidden layer.

10. How does the name counterpropagation signifies its architecture?

a) its ability to learn inverse mapping functions

b) its ability to learn forward mapping functions

c) its ability to learn forward and inverse mapping functions

d) none of the mentioned

Answer: c

Explanation: Counterpropagation network has ability to learn forward and inverse mapping functions.

This set of Neural Networks Multiple Choice Questions & Answers  focuses on “ART″.

1. An auto – associative network is?

a) network in neural which contains feedback

b) network in neural which contains loops

c) network in neural which no loops

d) none of the mentioned

Answer: a

Explanation: An auto – associative network contains feedback.

2. What is true about sigmoidal neurons?

a) can accept any vectors of real numbers as input

b) outputs a real number between 0 and 1

c) they are the most common type of neurons

d) all of the mentioned

Answer: d

Explanation: These all statements itself defines sigmoidal neurons.

3. The bidirectional associative memory is similar in principle to?

a) hebb learning model

b) boltzman model

c) Papert model

d) none of the mentioned

Answer: d

Explanation: The bidirectional associative memory is similar in principle to Hopfield model.

4. What does ART stand for?

a) Automatic resonance theory

b) Artificial resonance theory

c) Adaptive resonance theory

d) None of the mentioned

Answer: c

Explanation: ART stand for Adaptive resonance theory.

5. What is the purpose of ART?

a) take care of approximation in a network

b) take care of update of weights

c) take care of pattern storage

d) none of the mentioned

Answer: d

Explanation: Adaptive resonance theory take care of stability plasticity dilemma.

6. hat type learning is involved in ART?

a) supervised

b) unsupervised

c) supervised and unsupervised

d) none of the mentioned

Answer: b

Explanation: CPN is a unsupervised learning.

7. What type of inputs does ART – 1 receives?

a) bipolar

b) binary

c) both bipolar and binary

d) none of the mentiobned

Answer: b

Explanation: ART – 1 receives only binary inputs.

8. A greater value of ‘p’ the vigilance parameter leads to?

a) small clusters

b) bigger clusters

c) no change

d) none of the mentioned

Answer: a

Explanation: Input samples associated with same neuron get reduced.

9. ART is made to tackle?

a) stability problem

b) hard problems

c) storage problems

d) none of the mentioned

Answer: d

Explanation: ART is made to tackle stability – plasticity dilemma.

10. What does vigilance parameter in ART determines?

a) number of possible outputs

b) number of desired outputs

c) number of acceptable inputs

d) none of the mentioned

Answer: d

Explanation: Vigilance parameter in ART determines the tolerance of matching process.

This set of Neural Networks Multiple Choice Questions & Answers  focuses on “Applications Of Neural Networks – 1″.

1. Which application out of these of robots can be made of single layer feedforward network?

a) wall climbing

b) rotating arm and legs

c) gesture control

d) wall following

Answer: d

Explanation: Wall folloing is a simple task and doesn’t require any feedback.

2. Which is the most direct application of neural networks?

a) vector quantization

b) pattern mapping

c) pattern classification

d) control applications

Answer: c

Explanation: Its is the most direct and multilayer feedforward networks became popular because of this.

3. What are pros of neural networks over computers?

a) they have ability to learn b examples

b) they have real time high computational rates

c) they have more tolerance

d) all of the mentioned

Answer: d

Explanation: Because of their parallel structure, they have high computational rates than conventional computers, so all are true.

4. what is true about single layer associative neural networks?

a) performs pattern recognition

b) can find the parity of a picture

c) can determine whether two or more shapes in a picture are connected or not

d) none of the mentioned

Answer: a

Explanation: It can only perform pattern recognition, rest is not true for a single layer neural.

5. which of the following is false?

a) neural networks are artificial copy of the human brain

b) neural networks have high computational rates than conventional computers

c) neural networks learn by examples

d) none of the mentioned

Answer: d

Explanation: All statements are true for a neural network.

6. For what purpose, hamming network is suitable?

a) classification

b) association

c) pattern storage

d) none of the mentioned

Answer: a

Explanation: Hamming network performs template matching between stored templates and inputs.

7. What happens in upper subnet of the hamming network?

a) classification

b) storage

c) output

d) none of the mentioned

Answer: d

Explanation: In upper subnet, competitive interaction among units take place.

8. The competition in upper subnet of hamming network continues till?

a) only one unit remains negative

b) all units are destroyed

c) output of only one unit remains positive

d) none of the mentioned

Answer: c

Explanation: The competition in upper subnet of hamming network continues till output of only one unit remains positive.

9. What does the activation value of winner unit is indicative of?

a) greater the degradation more is the activation value of winning units

b) greater the degradation less is the activation value of winning units

c) greater the degradation more is the activation value of other units

d) greater the degradation less is the activation value of other units

Answer: b

Explanation: Simply, greater the degradation less is the activation value of winning units.

10. What does the matching score at first layer in recognition hamming network is indicative of?

a) dissimilarity of input pattern with patterns stored

b) noise immunity

c) similarity of input pattern with patterns stored

d) none of the mentioned

Answer: c

Explanation: Matching score is simply a indicative of similarity of input pattern with patterns stored.

This set of Neural Networks online test focuses on “Applications Of Neural Networks – 2”.

1. Can Invariances be build as static functions in the structure?

a) yes

b) no

Answer: b

Explanation: Invariances have to be dynamically estimated from data.

2. What is the objective of associative memories?

a) to store patters

b) to recall patterns

c) to store association between patterns

d) none of the mentioned

Answer: d

Explanation: The objective of associative memories is to store association between patterns for later recall of one of patterns given the other.

3. Is it possible to capture implicit reasoning process by patten classification network?

a) yes

b) maybe

c) no

d) cannot be determined

Answer: a

Explanation: For example neural network for contract bridge game.

4. Are classification methods based on correlation matching using moment features useful for problems of handwritten characters?

a) yes

b) no

Answer: b

Explanation: Because different parts of handwritten characters are deformed differently.

5. Associative memory, if used in feedback structure of hopfield type can function as?

a) data memory

b) cluster

c) content addressable memory

d) none of the mentioned

Answer: c

Explanation: Associative memory, if used in feedback structure of hopfield type can function as content addressable memory.

6. In feedforward network, the associations corresponding to input – output patterns are stored in?

a) activation state

b) output layer

c) hidden layer

d) none of the mentioned

Answer: d

Explanation: In feedforward network, the associations corresponding to input – output patterns are stored in weights of the network.

7. Which is one of the application of associative memories?

a) direct pattern recall

b) voice signal recall

c) mapping of the signal

d) image pattern recall from noisy clues

Answer: d

Explanation: The objective of associative memories is to store association between patterns for later recall of one of patterns given the other, so noisy versions of the same image can be recalled.

8. How can optimization be applied in images?

a) by use of simulated annealing

b) by attaching a feedback network

c) by adding an additional hidden layer

d) none of the mentioned

Answer: a

Explanation: Optimization be applied in images by use of simulated annealing to formulate the problem as energy minimization problem.

9. In control applications, how many ways are there to control a plant?

a) 1

b) 2

c) 4

d) infinite

Answer: b

Explanation: Open loop and feedback loop are the two ways.

10. Neuro – Fuzzy systems can lead to more powerful neural network?

a) yes

b) no

c) may be

d) cannot be determined

Answer: a

Explanation: If fuzzy logic is incorporated into conventional ANN models, more powerful systems can be created.