END TERM EXAMINATION [MAY-JUNE 2017] EIGHTH SEMESTER [B.TECH] SOFT COMPUTING [ETIT-410) Note: Attempt any five ‘Time: 3 Hrs. Max. Marks: 75 questions including Q. no. 1 which is compulsory. 1 (a) Differentiate between feed forward and feedback neural network. (5) Ans, [Architecture] Fa etorward newn Feed-back neural network | Layout Multiple layers of nodes ae Information flows in | including convolutional layers, different directions, simulating a memory effect. Data type Image data Sequence data i Input/ The size of the input and output | The size of the input and Output are fixed (i.e. input image with | output may vary (ie. fixed size and outputs the classification) receiving different texts and generating different translations for examples) Use cases | Image classification, recognition, | Text translation, natural | medical imagery, image analysis. | language processing, face detection. language translation, sentiment analysis, Drawbacks | Large training data Slow and complex training procedures, Description | CNN employs neuronal Time-series information is connection patterns. And they used by recurrent neural are inspired by the arrangement | networks. For instance, of the individual neurons in a user’s previous words the animal visual cortex, which | could influence the model allows them to respond to prediction on what he can overlapping areas ofthe visual | says next. field. Q.1. (b) Explain about fuzzy logics and its applications. (5) Ans. Fuzzy Logic is defined as a many-valued logic form which may have truth values of variables in any real number between 0 and 1. It is the handle concept of partial truth. In real life, we may come across a situation where we can’t decide whether the statement is true or false. At that time, fuzzy logie offers very valuable flexibility for reasoning. Fuzzy logic algorithm helps to solve a problem after considering all available data, Then it takes the best possible decision for the given the input. The FL method imitates the way of decision making in a human which consider al] the Possibilities between digital values ‘T and F Fuzzy logic applications ln automobiles, fuzzy logic is used for gear selection 4nd is based on factors such as engine load, réad conditions and style of driving —— a.2017-3 ersity-[B. ‘Tech|-Akash Books ry and power e washing strateg) f dishes and the level of food LP. Univ ‘ed to determine th eIn dishwashers, fuzzy logic is Us pa h as the number 0! needed, which is based on factors suc sidue on tl s anes ge based on factors such «In copy machines, fuzzy logic is used to adjust drum volta as humidity, picture density and temperature. ln sapave fuzzy logic is used to manage altitude control for satellites and spacecrafts based on environmental factors: vin medicine, fuzzy logic is used for compu such as symptoms and medical history. vin chemical distillation, fuzzy logic is used ¢ tor-aided diagnosis, based on factors o control pH and temperature variables «In natural language processing, fuzzy logic is used to determine semantic relations between concepts represented by words and other linguis' «In environmental control systems, such as air conditio logic determines output based on factors such as current temperature temperature. «Ina business rules engine, fuzzy logic may according to predetermined criteria. Q.1. (c) Define uncertainty and its usefulness in soft computing. (5) Ans. Uncertainty involved in any problem-solving situation is a result of some information deficiency. Information (pertaining to the model within which the situation is conceptualized) may be incomplete, fragmentary, not fully reliable, vague, contradictory, or deficient in some other way. Uncertainty analysis aims at quantifying the vari: to the variability of the input. The quantification is mos' statistical quantities of interest such as mean, median, estimation relies on uncertainty propagation techniques. Q.1. (d) Explain Genetic algorithm, why these algorithms are known as Genetic Algorithm. How is it usefully over simple traditional algorithm? (5) ‘Ans. Genetic Algorithm (GA) is a search-based optimization technique based on the s frequently used to find optimal principles of Genetics and Natural Selection. It i: or near-optimal solutions to difficult problems which otherwise would take a lifetime to solve. It is frequently used to solve optimization problems, in research, and in machine learning. Agenetic algorithm is a search heuristic that is inspired by Charles Darwin's theory of natural evolution. This algorithm reflects the process of natural selection where the fittest individuals are selected for reproduction in order to produce offspring of the next generation that’s why it is called Genetic algorithm. Advantages of GA over simple traditional algorithm Does not require any derivative information (which may not be available for many real-world problems) Is faster and more efficient as compared to the traditional methods. Has very good parallel capabilities. *Optimizes b c scret i-objecti ant a izes both continuous and discrete functions and also multi-objective *Provides a list of “good” solutions and not just a single solution. tic variables. ners and heaters, fuzzy and target be used to streamline decision-making ability of the output that is due often performed by estimating and population quantities. Thex &/ Righth Semester, Son Computing so sts an answer to the problem, whieh Bets better oy hee perenne: the search space 18 very large and there are large Po involved : parameters Replat Pereeptron Model with tho h ®xampy im - The Perceptron is a binary classifier which i input 5) veetor) to an output value fx) (a single binary value) Across the Matrix, (a "ebvalved fle) = {o if w, x +b>g 0 else where w is vector of real-valueg Weights a # weighted sum). b ig the ‘bias’, constant term ode Product (Which compy The value of fix) (9 To 1) is yy, © class; depend on “PY input val, instance, in the case ofa binary ¢| tion p, bl Cither a Positive or a \egative combination of MPuts must pro Positine bisn ative, then the Weighted the classifier neuron scat the 0 threshety S revit than | pu not the orientation) of t] decigi, e hot terminate if tj The >I in order to . i i i © decision bounda ate 7 cre tias ae he learning et is early is idereg the Simple, Q.2. (a) Explain and differentia Learning Paradi, een Upervised and Unsperviseg 7 (6.25) Ans. Refer Q.no. 2 ) End Te Exam 2018, Q2. (b) Explaj © Significance othidden layer. How itis useful in pattern recognition and contro} Pp (6.25) In neural networks, hidden layer is located between the input and o, of the algori 'm, in which the func through i i is a human face on a Picture, i hidden layers will break down ° uter has to detect all those eh image The nite Toews nPet image in order to identify features ane layers progressively get more abstract. ay level features such as edges while I ly connected layer with teurom ey etch ort ae ave the layers, we the end of eee! of counter propagation network, 625) Crassiication value is interpolative ay com were proposed by Heche Nielsen = Q3. (a ‘tion networl binations of the inpys, outa ty are mule ath do cnn ate eg oe ication of coun counter-propagation » ; 1987.They ar ‘The applicat ation. The cou in throe Ig ing layers. attern assoc 1. This model is neunt and clustering la ation and pal tstar mode an output approxim: instar-out faa’ prodaciog basically constructed PS inereeiiaet data mapping, basic: rforms ii rk that pet network r “tor y in . ‘he three }, + etitive learning. The threo |, in an the basis of comp : on t vector x, to inpu' response \ \LP. University-[B,Tech]|-Akash Books 2017-5 instar-outstar model are the input layer, the hidden (competitive) layer and the output layer. ‘There are two stages involved in t The input vector are clustered in the first stage. In the secon weights from the cluster layer units to the output units are tuned to obtain the d response. Q.3. (b) Explain the limitation of back propagation learning. Descrit Boltzmann machine. ‘Ans. The disadvantages of using a backpropagation algorithm are elt prefers a matrix-based approach over a mini-batch approach. he training process of a counter propagation net. id stage of training, the lesired be the (6.25) as follows: Data mining is sensitive to noise and irregularities Performance is highly dependent on input data. Training is time- and resource-intensive. Boltzmann Machine A Boltzmann machine is a type of recurren binary decisions with some bias. Boltzmann machines can be strung t more sophisticated systems such as deep belief networks. Some important points about Boltzmann Machine — They use recurrent structure. They consist of stochastic neurons, which have one of the two possible states, either 1 or 0. Some of the neurons in this are adaptive free state and some are clamped frozen state. elf we apply simulated annealing on discrete Hopfield network, then it would become Boltzmann Machine. Q.4. (a) Discuss the properties of fuzzy sets. Ans. Fuzzy sets follow some of the properties satisfied by crisp sets. In fact, crisp sets can be thought of as special instances of fuzzy sets. Any fuzzy set A is a subset of the refreshment X. Also, the membership of any element belonging to the null set 9 is 0 and the membership of any element belonging to the reference set is 1. The properties satisfied by fuzzy sets are: t neural network in which nodes make ogether to make (6.25) Commutativity: AvB=BUA AOB=BOA Associativity: Kv (Bue) = (AUB)UC An(Bae) = (AnB)oC Distributivity: Au (B00) = (AvB)o(Ave) Aa(Bus) = (AoB)L(AVC) Idempotence: AVA=A AAA Identity: Ave=t 62017 Righth Semester, Sof Computing AuX=q Aro=6 AvUX=x Transitivity: If A BCC. then 4 ¢ ¢ involution (Ay =% De Morgan's laws (AnJ’ = (ac - B) (Ay = (Atos) Since fuzzy sets can overlap, the laws of ry xcludeq Middle do Not hold Bood. Thus, AU Ktex ANA‘ #4 Q4 (b) Explain fuzzy rule feneration with examples? a Ans. Fuzzy Rule Generation 25) 1. apply fuzzy clusteri Ing X = fuzy partitio M matrix U . lu 2. use obtained U = lu.) to d y lefine membership functions * * usually X us multidimensional * assigning labels for one.. mains is easier -» 1. project U down to XypouX, 4X18, respectively 2. only consider Upper envelo, 3. linear interpolate membership values > mei 4 cylindrically extend membership functions * original clusters are interpreted as Conjunction of cy! extensions * ©g., cylindrical extensions “x, 18 low’, “x, is high” ~ multidimensional cluster label “x, is low and x, is high” * labeled clusters = classes characterized by labels * every cluster = one fuzzy rule Q.5. (a) Discuss and explain Crisp sets with its fun damental concept? (6.25) Ans. A set defined using a characteristic function that ssigns a value of either ing between members and mbership functions i 's “classical” or “ordinary” sets te laving propria oe anny manipulation ost e ivity: AUB=BUA vit Commutativity: 2 C=AU(BUC) ity (AUB)UC =AUIBU Associativity : aeenaee ri n a(AvC) vity: AVU(BOC) =(AU BAIA : oem A (BUC) =(ATB)U(ANC AUA=A Idempotence: 7Fe a Rg \ yy A Ye » LP. University-[B.Tech]-Akash Books 2017-7 Identify: AUO=A : ANE= ( An@= AVE Law of Absorption: AU(ANB) Ad(AUB) =A Transitivity: If A cB, Bc C then Ac C | Involution (AD =A . Law of the Excluded Middle: AU A= E Law of Contradiction: ADA ‘ De Morgan's laws: (AU By = ASO BE d (AN By = ASU BS j (6.25) Q5. (b) Discuss the operations performed on crisp relation? Ans. Operations performed on Crisp relation We will be using the following sets for further discussion: X=(1, 2, 3, 4,5, 6); A= (1, 2, 3}; B= (2,3, 4; C = (5, 6) Involution: Involution states that the complement of complemen For the given data, A’=X-A=(4,5, 6) (Ay =X-A’={1, 2,3)=A Commutativity: ‘The commutativity property states that the operation can be of the order of the operand. For example, addition is a commutative operator, t of set A would be set A itself. performed irrespective s02+3 or 3+ 2 yields the same result. But, subtraction is not commutative, so 3-24 2-3. Proving union is commutative: BoC=$o AvUB=(I, 2,3, 4} + LHS An (BNC) =$—> RHS BUA= (1, 2, 3,4] > RHS Distributivity: Proving intersection is Mathematically it is defined as, commutative: AB =(2,3} + LHS BOA =({2,3) + RHS Associativity: ‘The associativity property allows us to perform the operations by grouping the operands and keeping them in similar order. (AUB) UCsAuU (BUC) For given data: AUB=({I, 2,3, 4} (AV B)u C=I1, 2, 3, 4, 5, 6) + LHS. BUC = (2, 3, 4, 5, 6) AU(BUC)={I, 2,3, 4, 5, 6} » RHS (ANB) C=AN(BOC) For given data: AOB=(2,3) (ANB) C=6 LHS a AU (BOC) = (AU B)n (AU C) BaC=6 Au (BO C)={1, 2, 3}9LHS AUB={1,2,3, 4) AUC=(1, 2,3, 5,6) (AUB) (AU C)= (1, 2,3) > RHS An (BUC) (ANB) U (An ©) BUC= (2,3, 4, 5, 6) Aa (Bu C)= 12,3) > LHS AnB=12,3) AnC=6 (AN B)U(AN C) = (2, 3) > RHS Absorption: Mathematically absorption is defined as, AU(ANB)sA For the given data: AOB= (2,3)Eighth Semester, Soft Com, Puting 201 ; 8-20 AOB)=(1,2,3)=A For ASB ag | i iA B)=A a aa Siven data ‘the OBS For the given data iv Baitg gt AW B=(1, 2,3, 4) au rs i, art! Li / S i As(AUB)={1,2, "= t4 5, Idempotency/Tautology: a = (1,55) Idempotency is defined as, aBeie ALA=A ‘ ANA=A For the given data, AVA={1,2.3)— ADA=(1,2.3)24 Identity; Mathematically, we Property as, can define this erent learning (6.25) an example, (6.25) b) End Ter; Examinatio (a) Exp), he rious Properties of ithmetic and Lattice of fuzzy ny, ers, (6.25) Ans. Refer Q.no. 4a) End Term §, 'Mination 2018. Q7. (b) Dis, Hy yi) = 2. Intersection; The intersection of two fuzzy sets A and B is new fuzzy 4 B with membership function defined as Heal) = ay, Hyi LP. University-{B.Tech|~Akash Books 2017-9 set A with A membership function Hy) = 1-1, 4. Product of two fuzzy sets: The product of two fuzzy sets A and B is a new fuzzy A.B whose, membership function is defined as Hyasp (2) = Hy (x), byl) 5. Equality: Two fuzzy sets A and B are said to be equal (A = B) p, (x) = H,(). 6. Product of a fuzzy set with a crisp number: Multiplying a fuzzy set A by a crisp number result in a new fuzzy set product a. A with the membership function 1, , () =a. p(x). 7. Power of a fuzzy set: The « power of fuzzy set is a new fuzzy set A whose membership function is given by HA) = (ay(o))® Raising a fuzzy set to its second power is called concent (CON) taking the square rool is called Dilation (DIL). 8. Difference: The difference of two fuzzy sets N and B is a new fuzzy set A- B defined as A-B = (AnB) 9. Disjunctive SUM: The disjunctive sum of two fuzzy sets A and B isa fuzzy set A@B defined as A®B = (AnB)U(ANB) Q.8. Write short note on any two: Q8. (a) Associative Memo! Ans. Refer Q.no. 8(c) End Term Examination 2018. Q.8. (b) Hopfield’s Network Ans. Refer Q.no. 1 (c) End Term Examination 2018 Q.8. (c) Neuro Fuzzy System ~ ’ | 3 ( ' ft | 1 | / | | | | <4 | | t & | A |, Fig. Union of fuzzy sets Fig. Intersection of fuzzy sets 3. Complement: The complement of a fuzzy set Ais anew fuzzy set is a new fuzzy (6.25) Ans. The Neuro-fuzzy system is based on fuzzy system which is trained on the basis of the working of neural network theory. The learning process operates only on the local information and causes only local changes in the underlying fuzzy system. A neuro-fuzzy system can be seen as a S-layer feed forward neural network. The fret layer represents input variables, the middle (hidden) layer represents fuzzy rules and the third layer represents output variables. Fuzzy sets are encoded as connection weights within the layers of the network, whic h id ionality i i and training the model provides functionality in processingyoo Righth Semester, Soft Computing 10-2017 f oe +t I | put layer, each neuron transmits externa] Crisp si Working flow: In the in next layer. gnals directly to the * Each fuzzificati on neuron receives a crisp input and determines the degrag to which the input belongs to the input fuzzy set. * The fuzzy rule layer recei aggregation, *It has self-learning, self- izi elt can mimic the human decision-making Disadvantages; *Hard to develop a model from a fuzzy system, nding suitable membership *Neural networks cannot be Used if traini, Applications: * Student Modelling * Medical systems * Traffic contro} Systems ° Forecasting and Predictions Values for fuzzy Systems, ng data is not available.END TERM EXAMINATION [MAY-JUNE 2018] EIGHTH SEMESTER [B.TECH] ETIT-410] SOFT COMPUTING [ es Time: 3 Hrs. / Note: Attempt any five questions including Q. no- 1 which is Q.1. Attempt following in brief: Q.1. (a) Differentiate between hard and soft computing. (5) Ans. S.No. | Soft Computing 1 Soft Computing is liberal of compulsory. Hard Computing Hard computing needs a exactly | inexactness, uncertainty, partial state analytic model. L truth and approximation. 2. | Soft Computing relies on formal Hard computing relies on binary logic and probabilistic reasoning. logic and crisp system. 3. | Soft computing has the features of | Hard computing has the features approximation and dispositionality. | of exactitude(precision) and categoricity. 4 Soft computing is stochastic in Hard computing is deterministic nature. in nature. 5. _ | Soft computing works on ambiguous | Hard computing works on exact and noisy data. data. 6. Soft computing can perform parallel | Hard computing performs computations. sequential computations. 7. | Soft computing produces Hard computing produces precise approximate results, results. 8 Soft computing will emerge its own | Hard computing requires programs. programs to be written. EE A ~ [Hard com a re 7 9. | Soft computing incorporates Hard computing is settled. randomness . 10. | Soft computing will use multivalued | Hard computing uses two-valued logic. logic. __| Q.1. (b) Draw an architecture of Neural Network and explain. (5) Ans. Neural Network Architecture: Neural Networks are complex structures made of artificial neurons that can take in multiple inputs to produce a single output. This is the primary job of a Neural Network — to transform input into a meaningful output. Usually, a Neural Network consists of an input and output layer with one or multiple hidden layers within. It is also known as Artificial Neural Network or ANN. ANN architecture in Neural Network functions just like a human brain and is very important. , n't Neural Network, all the neurons influence each other, and hence, they are all connected. The network can acknowledge and observe every aspect of the dataset| | >. ee ~" oye ee Ee, ee — 2 awe ee “> © semen, —_ ews — a +o ~m < ve Se, ye te —_— Seca tS — se ~~ ee — * a pute > tien ee ete ee pee ene i tee te — at te ee g Came + tenes 5 ‘ a eS sag Mente te « ‘ a3 PRY Cerin ow , =? ey ~-f me « * VY?’ wy Cee ete a en alee peat he * SSS ee mee aa ~~. Pong *¢ ae & Gute tenn, ane © 10 ae, nateeiay SESS Se mf ee ee SoS: ae et then na ~ eee . tas, we sm corset tom, Sto a. —- .. eo — tine cane, cine, tw s = Ctr ne Ste Hem te ee oe ~*~. oo ee e . ee ~~ - : : nee ee ‘ * ¢ A oe + ee ee s : * wt se } mee fe.y Eighth Semester, Soft Computing 2-2018 +Feedforward Networks: In this model, the signals only ¢, el; towards the output layer Feedforward Networks have an input layers in ore layer with zero or multiple hidden layers. They are widely useq in pati. *Sing My "Feedback Networks: In this model, the recurrent or interagy NS their internal state (memory) to Process the sequence of inputs, Inve nent travel in both directions through the loops (hidden layer/s) in the tem, hy » Sip, Siac typically used in time-series and sequential tasks. etwony Bay Input layer hidden layer 1 hidden layer 2 output layer at is Hopfield network? Explain the algorithm to Store network. a *Weight/connection strength is represented by w,. ; °Connections can be excitatory as well as inhibitory. It would be excitatory, if th Output of the neuron is same as the input, otherwise inhibitory, + Weights should be symmetrical, i.e, wy=W, x 4 x ny % % Ya2-2018 Eighth Semester, Soft Co; *Feedforward Networks; ™Puting Input layer X hidden layer 1 hidden . at is Hopfielg Netwo; recall a set of bj layer 2 output layer Explain th ‘ipolar Patterns in Hy, The HoLP. University-[B Tech]-Akash Books 2018-3 Y, have the weights w,,, w, and w. t from Y to Y,, Y, and Lenape eee dal hts on them respectively. Similarly, other arcs have the weig) Training Algorithm During training of discrete Hop! that we can have the binary input vectors as well as bi th the cases, weight updates can be done with the following relation Case 1 - Binary input patterns For a set of binary patterns sp, p= 1toP Here, s P= 8; Ps 8) Pry 8, Pyoos8, Weight Matrix is given by field network, weights will be updated. As we know ipolar input vectors. Hence, in oP w= Y[2s\(p)=12s,(p)=11] fori ej = Case 2 - Bipolar input patterns For a set of binary patterns s p, p = 1 to P Here, sp =8, P, 8, Py---+ 8, Pym 8, P Weight Matrix is given by P w,= Y[s(plls,(p)] forizj mi Testing Algorithm Step 1 - Initialize the weights, which are obtained from training algorithm by sing Hebbian principle. Step 2 - Perform steps 3-9, if the activations of the network is not consolidated. Step 3 - For each input vector X, perform steps 4-8. Step 4 - Make initial activation of the network equal to the external input vector as follow: y,=%,fori=1ton Step 5 - For each unit Y,, perform steps 6-9. Step 6 - Calculate the net input of the network as follows- yn= 7 LIMs i Step 7 - Apply the activation as follows over the net input to calculate the output - Here 0, is the threshold. Step 8 - Broadcast this output y, to all other units. Step 9 - Test the network for conjunction. Q.1. (d) Explain the error correction process and gradient descent rule. (5) Ans. Error-Correction Learning, used with supervised learning, is the technique f comparing the system output to the desired output value, and using that error to irect the training. In the most direct route, the error values can be used to directly idjust the tap weights, using an algorithm such as the back propagation algorithm. If the system output is y, and the desired system output is known to be d, the error signal ‘an be defined as: e=d-yLP. University-[B Tech|-Akash Books 2018-3 ¥. and Y, have the weights w,,, ™,, and w,, ing to Y,, The output from Y, going * ie weights on them respectively. Similarly, other arcs have t Training Algorithm ; . 7 During training of discrete Hopfield network, weights will be updated ae , now that we can have the binary input vectors as well as bipolar input 3 . both the cases, weight updates can be done witl Case 1 - Binary input patterns For a set of binary patterns s p, p= 1 to P Here, s P= 8; Pr& Poor 8 Pron, P Weight Matrix is given by w= ¥ [25 (p)=12s,()-1] for ij pal fh the following relation Case 2 - Bipolar input patterns For a set of binary patterns s p, p = 1 to P Here, s p = 8, Py 8, Py 8 Pm 8, P Weight Matrix is given by P w,= D[s(alls,(p)] for ij pa Testing Algorithm Step 1 — Initialize the weights, which are obtained from training algorithm by sing Hebbian principle. Step 2 - Perform steps 3-9, if the activations of the network is not consolidated. Step 3 - For each input vector X, perform steps 4-8. Step 4 - Make initial activation of the network equal to the external input vector as follows — y,=x,fori=1ton Step 5 - For each unit Y,, perform steps 6-9. Step 6 - Calculate the net input of the network as follows- yn = it LMy 7 Step 7 - Apply the activation as follows over the net input to calculate the output - Here 0, is the threshold. Step 8 - Broadcast this output y, to all other units. Step 9 - Test the network for conjunction. Q.1. (d) Explain the error correction process and gradient descent rule. (5) Ans. Error-Correction Learning, used with supervised learning, is the technique f comparing the system output to the desired output value, and using that error to irect the training. In the most direct route, the error values can be used to directly just the tap weights, using an algorithm such as the back propagation algorithm, If 1e system output is y, and the desired system output is known to be d, the ignal an be defined as: , the error sign: e=d-yemcee aa aes 2018 Kighth Semester, Soft Computing Error correction learning algorithms attempt to Minimize this erro, training iteration Sign, ‘The gradient descent algorithm is not specifically an ANN lear has « large variety of uses in various fields of science, engineering However, we need to discuss the gradient descent algorithtn in order vy na” the back propagation algorithm, The gradient descent algorith ® error function g(y), through the manipulation of a " Wn ig Us weight vector wt should be a linear combination of the weight vector a a ind an input Vector y wy in + 1] w, ln] + new, In}) Here, 1 is known as the step. ve parameter, and affects the rat, of the algorithm. If the step size is too small, the algorithm will tags" Mvergen,, converge. If the step size is too large the algorithm might oscilla The gradient descent algorithm works by taking the gradi to find the path of steepest descent. By following the path of steepes, dean i8* apa ‘eration, we will either find a minimum, or the algorithm co : “pace is infinitely decreasing. When a minimum is found, there j : , there is 7 Rh 1s a global minimum, however, RO Buarantes thay vd Q.1. (©) Find A sets: A= (1/10 + 0.75/15 + 0.3/2.0 +0, B= (1/1.0+ 0.6/1.5 + 0.22.0 40,1 Ans. AUB = lent of ~ Band complement of AW B for the following two tug, o 29 a 15/2.5 + 0/3) /2.5 + 0/3} max{M q(x), My(x)} vm 0.15 0 1 (0.75 03 Tote ee 25° 3 s+ 10 15 "26 Ie Q.2. (a) Differentiate between supervised and unsupervised learning. Give one example of each. (6) Ans. Refer Q. No. 3 First Term Exam 2017 22, (b) Describe McCulloch-Pitts Neuron. Implement “AND” function using McCulloch-Pitts Neuron. (6.5) Ans. ‘The idea of the simple neuron model first emerged in the 1940s with the work of McCulloch and Pitts, The cybernetics movement psychology, engineering and mathematics suiting in neurons which would perform a number of: neural networks developing internal representation re put forward the theory of lated to experience aaLP. University-{B-Tech|-Akash Books 2018-6 Weights Activation Input (fixed) ci Output a signal ay WwW Output UM Summing part function f() Fig. McCulloch and Pits Model of Neuron In the 1950s, research continued initially into the development of networks to form specific tasks but this changed and the goal became to develop machines at could learn. By the end of that decade there here had been a lack of significant -velopments and work in this field diminished considerably. ‘The perceptron effectively splits the input patterns into two distinct regions with e region being represented by a 1 on the output and the other a 0. Rosenblatt’s training gorithm for the perceptron would converge if the input patterns to the perceptron were early separable. The perceptron would therefore approximate the decision boundary tween the two classes of outputs. Perceptrons were successfully trained to perform certain tasks but there were ilures that could not be overcome, Minsky and Papert pointed out the serious shortcomings of perceptrons and interest in the study of neural networks again declined, ‘The ‘AND’ function using McCulloch-Pits neuron is shown below y=, AND x, Xx, =Oor! % x)= 00r1 AND gate018 Eighth Semester, Soft Computing 6-21 Q.3. (a) What are activation function? Differentiate Sigmodial and Bipolar Sigmoidal function. Ans. The activation function is used to calculate the output Tespongg . The sum of the weighted input signal is applied with an activation to obtain, . a Me, , For neurons in same layer, same activation functions are used. There may rey n well as nonlinear activation functions. The nonlinear activation functions = ding, multilayer network ; Uses Binary Sigmoidal Function This is also called logistic function. It ranges between 0 and 1. betwee, : iy, 1 fix) = log sig (x) = Trexp™ +exp If fix) is differentiated we get, flx) = flx) [1- fla) Fig. Shows the binary sigmoidal function, 10 = os oe 07 06 o———— oe 09 02 a ° 4s E200 49444 ete sere se Fig. Binary Sigmoidal Functions s The desired range here is between +1 and -1. The function is related to the hyperbolic tangent function, ‘The bipolar sigmoidal function is given as, yx) = 2flx)-1 Substituting the value of flx) we get, Fig. Bipolar Sigmoidal Function tput of the neuron Y for the network shown in inary Sigmodial and Bipolar Sigmoidal function.(6.5) 08 10 —+@~ ot 035 Hig following figure using BiLP. University-[B.Tech|~Akash Books 2018-7 Ans. The given network has three input neurons with bias and one output neuron, These form a layer network, The inputs are given as, \x,, x), x] = (0.8,0.6,0.4] The weights are, (w,, Wy, Wa] = (0.1, 0.3, - 0.2] The net input can be calculated as, Yn = b+ DY" (yw) Yi, = 0-35 + 0.8 x 0.1 + 0.6 x 0.3 + 0.4 x (0.2) Y,, = 0.35 + 0.08 + 0.18 - 0.08 = 0.53 @ For Binary Sigmoidal Function, 1. . = fly,)= ae Topue = 062 (ii) For Bipolar Sigmoidal activation a 1-e% y = fly,)= -1= at +e 1+e% 1-e° y = let" 20.259 if a Q.4. (a) What are Fuzzy Set? Enlist and explain various operators on Fuzzy . What do you mean by Lambda-Cut? (6) Ans. Fuzzy sets support a flexible sense of membership of elements to a set. A fuzzy t is defined as follows: If X is universe of discourse and x is a particular element of X, then a fuzzy set. A fined on X may be written as a collection of ordered pairs, = {suge, xix} where each pair (x, 4,(x) is called a singleton. Operations on Fuzzy set are: (i) Ba gle) = max(yj (x), 4g (x) Gi) wg gla) = minty (x),ng (2) Gi) pyle) = 1-4, g(x) (iv) jg) =H dng (a) (W) Hy ge) = ang (x) (i) ppg a) = (ug (a))* (i) A-B=(AnBS) (viii) A @ B= (An B)U(AnB')eeeeeowo eo ; 8-2018 Kighth Semester, Soft Computing Q-4. (b) What is fuzzy relation? Draw a bipartite 8nd 5) the following relation X = (X1, X2, X3, X4) pl, Ze, xl x2 x3 x4 & (02 0 05 9 - %/0 03 0.7 og afc ets, which ar zy relation is based on the concept that or unrelated. A fuzzy relation is a tne, fuzzy set defined on the Cartesian Product of X,,.. X,) where tuples (X).--) X,) may have varying degrees of p(x x “lat relation, That is, ener N wig RAK Ke f Meta) [Ca 95.3), *,€X, i XpXpk. aX, A fuzzy relation between two sets X and Y ig called binary fuz, Telat denoted by R(X, Y). A binary relation R(X, Y) is referred te a8 bipartite got X + Y. The binary relation on a single set X is cai relation occurs when X=Y H 4) consed iC 2 Fig. Simple fuzzy graphLP. University-[B.Tech|-Akash Books 2018-9 Q5. (a) What is defuzzification method? Enlist and explain various zzification methods. (6) Ans. Refer Q.no. 5 (a) of End Term Exam 2018 Q.5. (b) What is fuzzy inference system? Explain all types of is fuzzy rence system. What is fuzzy preposition? (6.5) Ans. Fuzzy inference is the process of formulating the mapping from a given input an output using fuzzy logic. The mapping then provides a basis from which decisions yn be made, or patterns discerned. Following are the two important methods of FIS, having different consequent of y rules — *Mamdani Fuzzy Inference System *Takagi-Sugeno Fuzzy Model (TS Method) Mamdani Fuzzy Inference System This system was proposed in 1975 by Ebhasim Mamdani. Basically, it was ticipated to control a steam engine and boiler combination by synthesizing a set of zy Tules obtained from people working on the system. Steps for Computing the Output Following steps need to be followed to compute the output from this FIS - Step 1 - Set of fuzzy rules need to be determined in this step. Step 2 - In this step, by using input membership function, the input would be de fuzzy Step 3 - Now establish the rule strength by combining the fuzzified inputs according fuzzy rules. Step 4 - In this step, determine the consequent of rule by combining the rule ength and the output membership function. Step 5 - For getting output distribution combine all the consequents. Step 6 - Finally, a defuzzified output distribution is obtained. Following is a block diagram of Mamdani Fuzzy Interface System sseengh IF — THEN outa Jj: {| Takagi-Sugeno Fuzzy Model (TS Method) This model was proposed by Takagi, Sugeno and Kang in 1985, Format of this rule given as - ee | = me IF xis A and y is B THEN Z = fix, y) Here, AB are fuzzy sets in antecedents and z = fix, y) is a crisp function in the mnsequent.i. > ‘| 10-2018 Kighth Semester, Soft Computing Fuzzy Inference Process The fuzzy inference process under Takagi-Sugeno Fuzy zy M. in the following way del (TS My Step 1: Fuzzifying the inputs — Here, Step 2: Applying the fuzzy operator applied to get the output. how, the inputs of the Syste, ~ In this ste fi Te mag m this step, the fuzzy operas Q.6. (a) What are genetic algorithms? How Mutation, Crossover works in genetic algorithms? Explain, ” Selection Ans. Refer Q.no. 1 (f) of End Term Exam 2018. ; Q.6. (b) What are linguistic variables? How they are different fr variable. °™ Dame, Ans. Refer Q.no. 3 (a) of First Term Exam 2018. 6 Q.7. (a) What is learning in neural networks? Explain linear Sepa, non-linearly separable pattern with example. Table Ans. Learning, in artificial neural network, is the method of modifying the oF connections between the neurons of a specified network. Learning rs ANN classified into three categories namely supervised learning, uncupervecy learning reinforcement learning. less We say they're separable if there's a classifier whose decision boundary separ the positive objects from the negative ones. If such a decision boundary is a lin, ‘anction of the features, we say that the classes are linearly separable For example, consider a dataset with two fe: (-1,-),(, 1), to the other. atures x, and x, in which the poin (-3, -3),(4, 4) belong to one class and (~ 1, DC, - DJS 5, 2),(4, ~ 8) belog Asset of input vectors (or a training hyperplane exists such that each vect Q.7. (b) Explain error back of flowchart. . set) will be said to be linearly non-separable if tor lies on the pre-assigned side of the hyperplan propagation training algorithm with the hel 65 Ans. Step 1: Normalized the inputs and outputs with respect to their maximur values. It is proved that the neutral networks work better if input and outputs li r), between 0-1. For each training pair, assume there are ‘? inputs given by e and outputs 4 in a normalized form, nx Step 2: Assume the number of neurons in the hidden layer to He between /