{"pinia":{"ui":{"prefs":{"theme":"system","textScale":"md","fontFamily":"sans","reducedMotion":false,"highContrast":false}},"academy":{"universities":[],"universitySummaries":[{"slug":"rtu","name":"RTU","fullName":"Rajasthan Technical University","branches":[{"slug":"first-year-common","name":"First Year (Common)"},{"slug":"computer-science","name":"Computer Science"},{"slug":"mechanical","name":"Mechanical"},{"slug":"civil","name":"Civil"},{"slug":"ai-data-science","name":"AI & Data Science"},{"slug":"ee-ec-eee","name":"EE / EC / EEE"}]},{"slug":"aktu","name":"AKTU","fullName":"Dr. A. P. J. Abdul Kalam Technical University","branches":[{"slug":"first-year-common","name":"First Year (Common)"},{"slug":"computer-science","name":"Computer Science"}]}],"activePapers":[{"id":"soft-computing-2023-8E1850","title":"Soft Computing (June 2023)","subjectSlug":"soft-computing","subjectName":"Soft Computing","universitySlug":"rtu","branchSlug":"ee-ec-eee","year":4,"semester":8,"paperCode":"8E1850","examYear":2023,"questions":[{"id":"soft-computing-2023-8E1850-pA-q1","number":1,"marks":2,"tags":["Part A"],"text":[{"type":"paragraph","content":"Q.1. Describe the importance of the Back Propagation Network."}],"answer":{"explanation":[{"type":"paragraph","content":"The Back Propagation Network is important because it provides a systematic, gradient-descent-based method for training multi-layer feed-forward neural networks by propagating the output error backward through the network layers, enabling the network to learn complex, non-linear input-output mappings that a single-layer perceptron cannot represent, and it remains the foundational training algorithm underlying most modern deep learning architectures."}]}},{"id":"soft-computing-2023-8E1850-pA-q2","number":2,"marks":2,"tags":["Part A"],"text":[{"type":"paragraph","content":"Q.2. Write three application scopes of the Neural Network."}],"answer":{"explanation":[{"type":"list","items":["Pattern recognition and classification, such as handwritten character recognition and image classification.","Function approximation and time-series prediction, such as stock price or load forecasting.","Control systems and robotics, such as adaptive process control and autonomous navigation."]}]}},{"id":"soft-computing-2023-8E1850-pA-q3","number":3,"marks":2,"tags":["Part A"],"text":[{"type":"paragraph","content":"Q.3. With the help of an example, state the role of bias in determining the net output of an Artificial Neural Network."}],"answer":{"explanation":[{"type":"paragraph","content":"Bias is an additional constant input (weight times a fixed input of 1) added to the weighted sum before the activation function, shifting the activation function's threshold. For example, in net = w1x1 + w2x2 + b, the bias b allows the neuron to output a non-zero value even when x1=x2=0, shifting the decision boundary away from the origin and giving the network greater flexibility to fit the training data."}]}},{"id":"soft-computing-2023-8E1850-pA-q4","number":4,"marks":2,"tags":["Part A"],"text":[{"type":"paragraph","content":"Q.4. What is Activation Function and write its importance."}],"answer":{"explanation":[{"type":"paragraph","content":"An activation function is a mathematical function (such as sigmoid, tanh, or ReLU) applied to a neuron's weighted input sum to produce its output. Its importance lies in introducing non-linearity into the network, without which any multi-layer network would collapse mathematically into an equivalent single-layer linear model, incapable of learning complex, non-linear patterns."}]}},{"id":"soft-computing-2023-8E1850-pA-q5","number":5,"marks":2,"tags":["Part A"],"text":[{"type":"paragraph","content":"Q.5. List out the steps in the perceptron learning algorithm for single output classes."}],"answer":{"explanation":[{"type":"list","items":["Initialize weights and bias (commonly to zero or small random values) and set the learning rate.","Present a training input pattern and compute the net input and output using the activation function.","Compare the computed output to the target output; if they differ, update the weights and bias using the perceptron learning rule.","Repeat over all training patterns until no weight changes occur (convergence) or a maximum epoch limit is reached."]}]}},{"id":"soft-computing-2023-8E1850-pA-q6","number":6,"marks":2,"tags":["Part A"],"text":[{"type":"paragraph","content":"Q.6. Compare supervised and unsupervised learning approaches in ANN."}],"answer":{"explanation":[{"type":"paragraph","content":"Supervised learning trains the network using labeled input-output pairs, adjusting weights to minimize the error between actual and desired output (as in backpropagation). Unsupervised learning trains the network using only input data with no target output, allowing the network to discover inherent structure or clusters in the data on its own (as in self-organizing maps)."}]}},{"id":"soft-computing-2023-8E1850-pA-q7","number":7,"marks":2,"tags":["Part A"],"text":[{"type":"paragraph","content":"Q.7. Consider the following binary strings (101010110) and (100110100). Write the new strings using a single-point crossover operator and by choosing the 3rd site at random and mutating the 4th bit of the string."}],"answer":{"explanation":[{"type":"paragraph","content":"Parents: 101010110 and 100110100. Crossover at site 3 (first 3 bits fixed, remainder swapped) gives offspring 101|110100 = 101110100 and 100|010110 = 100010110. Mutating the 4th bit of each (flipping it) gives final strings 101010100 and 100110110."},{"type":"math","content":"\\text{Offspring 1: } 101010100, \\quad \\text{Offspring 2: } 100110110"}]}},{"id":"soft-computing-2023-8E1850-pA-q8","number":8,"marks":2,"tags":["Part A"],"text":[{"type":"paragraph","content":"Q.8. In PSO, if the value of C1 and C2 is 1.5, the values of random numbers R1 and R2 are 0.5, Xgbest is 3.5, Xpbest is 2.5, X(i) is 2.5 and V(i) is 2.0 then what will be the new position of X(i)."}],"answer":{"explanation":[{"type":"paragraph","content":"New velocity V(i+1) = V(i) + C1*R1*(Xpbest - X(i)) + C2*R2*(Xgbest - X(i)) = 2.0 + 1.5(0.5)(2.5-2.5) + 1.5(0.5)(3.5-2.5) = 2.0 + 0 + 0.75 = 2.75. New position X(i+1) = X(i) + V(i+1) = 2.5 + 2.75 = 5.25."},{"type":"math","content":"X_{new} = 2.5 + 2.75 = 5.25"}]}},{"id":"soft-computing-2023-8E1850-pA-q9","number":9,"marks":2,"tags":["Part A"],"text":[{"type":"paragraph","content":"Q.9. Define an artificial neural network. Draw its mathematical model."}],"answer":{"explanation":[{"type":"paragraph","content":"An artificial neural network is a computational model inspired by biological neurons, consisting of interconnected processing units (neurons) organized in layers, each connection carrying a weight, that collectively learn to map inputs to outputs through weight adjustment."},{"type":"math","content":"y = f\\left(\\sum_{i=1}^n w_i x_i + b\\right)"}]}},{"id":"soft-computing-2023-8E1850-pA-q10","number":10,"marks":2,"tags":["Part A"],"text":[{"type":"paragraph","content":"Q.10. State the concept of delta-rule used in Adaptive Linear Neurons."}],"answer":{"explanation":[{"type":"paragraph","content":"The delta rule (Widrow-Hoff rule) updates weights in proportion to the error between the target and the actual net (linear) output, before applying any activation function, using gradient descent to minimize the mean squared error, forming the basis of the Adaline (Adaptive Linear Neuron) learning algorithm."}]}},{"id":"soft-computing-2023-8E1850-pB-q1","number":1,"marks":8,"tags":["Part B"],"text":[{"type":"paragraph","content":"Q.1. Given two universes X={x1, x2, x3, x4, x5} and Y={y1, y2, y3, y4, y5}, the fuzzy sets A defined on X and fuzzy set B defined on Y are given below. (i) Find the relation R = A x B, where A = {0.4/x1 + 0.7/x2 + 1/x3 + 0.8/x4 + 0.6/x5}, B = {0.2/y1 + 0.6/y2 + 1/y3 + 0.9/y4 + 0.7/y5}. Consider another fuzzy set C defined on the universe V={v1, v2, v3}, C = {0.4/v1 + 1/v2 + 0.8/v3}. (ii) Find P = B x C. (iii) Using max-min composition find R o P."}],"answer":{"explanation":[{"type":"paragraph","content":"Given A = {0.4/x1, 0.7/x2, 1/x3, 0.8/x4, 0.6/x5}, B = {0.2/y1, 0.6/y2, 1/y3, 0.9/y4, 0.7/y5}, and C = {0.4/v1, 1/v2, 0.8/v3}. The fuzzy Cartesian product (relation) R = A x B is computed element-wise as R(xi,yj) = min(A(xi), B(yj))."},{"type":"table","headers":["R","y1(0.2)","y2(0.6)","y3(1.0)","y4(0.9)","y5(0.7)"],"rows":[["x1(0.4)","0.2","0.4","0.4","0.4","0.4"],["x2(0.7)","0.2","0.6","0.7","0.7","0.7"],["x3(1.0)","0.2","0.6","1.0","0.9","0.7"],["x4(0.8)","0.2","0.6","0.8","0.8","0.7"],["x5(0.6)","0.2","0.6","0.6","0.6","0.6"]]},{"type":"paragraph","content":"Similarly, P = B x C is computed as P(yj,vk) = min(B(yj), C(vk))."},{"type":"table","headers":["P","v1(0.4)","v2(1.0)","v3(0.8)"],"rows":[["y1(0.2)","0.2","0.2","0.2"],["y2(0.6)","0.4","0.6","0.6"],["y3(1.0)","0.4","1.0","0.8"],["y4(0.9)","0.4","0.9","0.8"],["y5(0.7)","0.4","0.7","0.7"]]},{"type":"paragraph","content":"Finally, R o P is computed using max-min composition: (RoP)(xi,vk) = max over j of min(R(xi,yj), P(yj,vk)). For example, for x1 and v2: min(R(x1,y1),P(y1,v2))=min(0.2,0.2)=0.2; min(R(x1,y2),P(y2,v2))=min(0.4,0.6)=0.4; min(R(x1,y3),P(y3,v2))=min(0.4,1.0)=0.4; min(R(x1,y4),P(y4,v2))=min(0.4,0.9)=0.4; min(R(x1,y5),P(y5,v2))=min(0.4,0.7)=0.4; maximum of these is 0.4."},{"type":"table","headers":["RoP","v1","v2","v3"],"rows":[["x1","0.4","0.4","0.4"],["x2","0.4","0.7","0.7"],["x3","0.4","1.0","0.8"],["x4","0.4","0.8","0.8"],["x5","0.4","0.6","0.6"]]}]}},{"id":"soft-computing-2023-8E1850-pB-q2","number":2,"marks":8,"tags":["Part B"],"text":[{"type":"paragraph","content":"Q.2. Let A be a fuzzy set that tells about a student as shown in the figure, where the linguistic variable P represents a Pass student, F stands for a Fair student, G represents a Good student, and VG represents a Very Good student. Triangular membership functions are shown with peaks approximately: P peak 0.6 at x=60, F peak 0.4 at x=70, VG and V peaks approximately 0.2 at x=80 and x=90, over the universe of discourse x=50 to 100. Calculate the defuzzified value of the fuzzy set A with weighted average method and center of sums."}],"answer":{"explanation":[{"type":"paragraph","content":"Reading the given triangular membership plot, the peak (representative) points of each linguistic term and their corresponding membership heights are approximately: P at x=60 with mu=0.6, F at x=70 with mu=0.4, and VG and V at x=80 and x=90 respectively, each with mu=0.2 (these are the illustrative values read from the described figure; in an actual examination, the candidate reads these directly off the printed graph)."},{"type":"subheading","content":"Weighted Average Method"},{"type":"paragraph","content":"The weighted average method defuzzifies by taking the sum of each peak value multiplied by its membership degree, divided by the sum of the membership degrees: x* = sum(mu_i * x_i) / sum(mu_i)."},{"type":"math","content":"x^* = \\frac{(0.6)(60) + (0.4)(70) + (0.2)(80) + (0.2)(90)}{0.6+0.4+0.2+0.2} = \\frac{36+28+16+18}{1.4} = \\frac{98}{1.4} = 70"},{"type":"subheading","content":"Center of Sums Method"},{"type":"paragraph","content":"The center of sums method is similar to the weighted average but uses the area (or, for simple triangular approximations without overlap correction, again the peak value times height as a simplified proxy for area) contributed by each fuzzy set's membership function, summed over all the sets, without subtracting overlapping regions counted twice (unlike the center of gravity method, which would require integrating the true union membership function). Using the same peak-height values as areas (a common simplification when only peak heights, not full triangle base widths, are given), the center of sums calculation gives the same numerical result of x* = 70 in this specific case, since both methods reduce to the same weighted-average-style computation when only representative peak points (rather than full membership function shapes) are used as the basis of calculation."},{"type":"math","content":"x^*_{cos} = \\frac{\\sum \\mu_i x_i}{\\sum \\mu_i} = 70"},{"type":"paragraph","content":"It should be noted that the exact numerical answer depends on the precise peak coordinates and heights read directly from the given figure in the original examination paper; the method demonstrated here, computing a mu-weighted average of the representative x-values for each linguistic term, is the standard, generally applicable procedure for both weighted average and (in this simplified peak-only form) center of sums defuzzification, and should be applied using the specific numerical values visible on the actual printed graph."}]}},{"id":"soft-computing-2023-8E1850-pB-q3","number":3,"marks":8,"tags":["Part B"],"text":[{"type":"paragraph","content":"Q.3. (a) With the help of examples, explain the various fuzzy propositions. [4]"},{"type":"paragraph","content":"(b) Explain the different methods for fuzzy approximate reasoning. [4]"}],"answer":{"explanation":[{"type":"subheading","content":"Fuzzy Propositions"},{"type":"paragraph","content":"A fuzzy proposition is a statement involving a linguistic variable and a fuzzy set, assigned a truth value between 0 and 1 rather than a strict binary true/false, unlike a classical (crisp) proposition. Unconditional and unqualified fuzzy propositions have the form 'x is A', such as 'the temperature is high', where the truth value equals the membership degree of x in fuzzy set A. Unconditional but qualified propositions add a truth or probability qualifier, such as 'x is A is very true' or 'x is A is likely'. Conditional and unqualified propositions have the form 'if x is A then y is B', such as 'if the temperature is high, then the fan speed is fast', forming the basis of fuzzy if-then rules used in fuzzy inference systems. Conditional and qualified propositions combine both a conditional structure and a qualifier, such as 'if x is A then y is B is very true'."},{"type":"subheading","content":"Methods for Fuzzy Approximate Reasoning"},{"type":"paragraph","content":"Fuzzy approximate reasoning (fuzzy inference) draws conclusions from fuzzy if-then rules and fuzzy facts, even when the given facts do not exactly match the rule's antecedent. The generalized modus ponens method extends classical modus ponens: given rule 'if x is A then y is B' and fact 'x is A-prime' (similar to but not identical to A), it concludes 'y is B-prime' using a fuzzy implication relation and the compositional rule of inference (max-min composition of A-prime with the fuzzy relation representing the rule). Mamdani-style inference uses min for implication and max for aggregation of multiple rules, producing an output fuzzy set that is then defuzzified. Sugeno-style inference instead uses rule consequents that are linear functions or constants of the inputs rather than fuzzy sets, simplifying defuzzification to a weighted average calculation. Compositional rule of inference is the general mathematical mechanism (using max-min or max-product composition) underlying both of these specific inference styles, formally deriving the output fuzzy set from the input fuzzy set and the rule's fuzzy relation."}]}},{"id":"soft-computing-2023-8E1850-pB-q4","number":4,"marks":8,"tags":["Part B"],"text":[{"type":"paragraph","content":"Q.4. Explain the Associative Neural Memory Models, Adaptive Resonance Theory and various applications of ANN."}],"answer":{"explanation":[{"type":"subheading","content":"Associative Neural Memory Models"},{"type":"paragraph","content":"Associative memory networks store a set of pattern pairs (or single patterns) such that presenting a complete or partial (noisy) version of an input pattern retrieves the associated stored pattern. Auto-associative memory (such as the Hopfield network) stores patterns that map to themselves, useful for pattern completion and noise removal, using a fully interconnected, symmetric-weight recurrent network that settles into a stable state corresponding to the nearest stored pattern. Hetero-associative memory stores pairs of distinct patterns (input pattern A associated with a different output pattern B), useful for pattern association tasks such as translating one representation into another."},{"type":"subheading","content":"Adaptive Resonance Theory (ART)"},{"type":"paragraph","content":"Adaptive Resonance Theory is an unsupervised neural network architecture designed to solve the stability-plasticity dilemma, allowing the network to learn new patterns (plasticity) without catastrophically forgetting previously learned patterns (stability). ART networks compare an input pattern against stored prototype categories using a vigilance parameter that controls how closely a new input must match an existing category before being assigned to it; if no sufficiently close match is found, a new category is created rather than distorting an existing one, giving ART networks the ability to continue learning new categories indefinitely without degrading previously learned knowledge."},{"type":"subheading","content":"Applications of ANN"},{"type":"paragraph","content":"Neural networks are applied across pattern recognition (character and speech recognition), forecasting (load and financial forecasting), control systems (adaptive and non-linear process control), medical diagnosis (disease classification from clinical data), image processing (object detection and classification), and optimization problems (combined with other soft computing techniques such as genetic algorithms), owing to their ability to learn complex, non-linear relationships directly from data without requiring an explicit analytical model of the underlying system."}]}},{"id":"soft-computing-2023-8E1850-pB-q5","number":5,"marks":8,"tags":["Part B"],"text":[{"type":"paragraph","content":"Q.5. Explain different defuzzification methods."}],"answer":{"explanation":[{"type":"paragraph","content":"Defuzzification converts a fuzzy output set (produced by a fuzzy inference engine) back into a single crisp numerical value suitable for practical use, such as a control action."},{"type":"list","items":["Centroid (center of gravity) method: computes the x-coordinate of the centroid of the area under the aggregated output membership function, x* = integral(x*mu(x)dx)/integral(mu(x)dx), giving the most commonly used and generally most accurate defuzzification method, though computationally the most intensive.","Weighted average method: computes a mu-weighted average of representative points (such as the peak of each contributing fuzzy set), x* = sum(mu_i * x_i)/sum(mu_i), computationally simpler than the full centroid method and widely used when output membership functions are symmetric.","Center of sums method: similar to the weighted average but sums the individual areas (or peak-weighted contributions) of all rule outputs without correcting for overlapping regions, making it computationally faster than centroid but potentially less accurate when rule outputs overlap significantly.","Mean of maxima method: takes the average of all x-values at which the aggregated membership function attains its maximum value, ignoring the overall shape of the membership function elsewhere, computationally very simple but can produce discontinuous control action as input conditions change smoothly.","Smallest/largest of maxima method: selects the smallest or largest x-value at which the maximum membership occurs, used in specific applications where a conservative (smallest) or aggressive (largest) control action is specifically desired."]}]}},{"id":"soft-computing-2023-8E1850-pB-q6","number":6,"marks":8,"tags":["Part B"],"text":[{"type":"paragraph","content":"Q.6. (a) With the help of a block diagram explain Genetic Neuro Hybrid Systems. [4]"},{"type":"paragraph","content":"(b) Write a note on Fuzzy Qualifiers. [4]"}],"answer":{"explanation":[{"type":"subheading","content":"Genetic Neuro Hybrid Systems"},{"type":"paragraph","content":"Genetic-neuro hybrid systems combine genetic algorithms with artificial neural networks, most commonly using the genetic algorithm to optimize the neural network's architecture (number of layers and neurons) or its connection weights, exploiting the genetic algorithm's global search capability to avoid the local minima that gradient-based training methods such as backpropagation can become trapped in."},{"type":"svg","alt":"Block diagram of genetic neuro hybrid system showing genetic algorithm optimizing neural network weights or structure evaluated against training data","content":""},{"type":"paragraph","content":"In this arrangement, each candidate solution (chromosome) in the genetic algorithm's population encodes a specific neural network configuration (such as a vector of connection weights, or a structural encoding of the number of hidden neurons), and the fitness of each chromosome is evaluated by constructing the corresponding neural network and measuring its performance (such as classification accuracy or mean squared error) on a training or validation dataset; the genetic algorithm then applies selection, crossover, and mutation to evolve progressively better-performing network configurations across successive generations."},{"type":"subheading","content":"Fuzzy Qualifiers"},{"type":"paragraph","content":"Fuzzy qualifiers modify a base fuzzy proposition to express an additional degree of truth, probability, or possibility, rather than simply asserting the proposition outright. Truth qualifiers (such as 'very true', 'fairly true', 'false') modify the certainty with which a proposition 'x is A' is asserted, effectively transforming the membership function through a modifier operation. Probability qualifiers (such as 'likely', 'unlikely') express the likelihood of a fuzzy event occurring. Possibility qualifiers express the degree to which a proposition is possible rather than certain. Qualified propositions allow fuzzy logic to express nuanced, hedge-modified natural language statements (such as 'it is very likely that the temperature is high') that more closely reflect natural human linguistic expression than simple, unqualified fuzzy propositions alone."}]}},{"id":"soft-computing-2023-8E1850-pB-q7","number":7,"marks":8,"tags":["Part B"],"text":[{"type":"paragraph","content":"Q.7. Explain the various applications of Fuzzy Sets."}],"answer":{"explanation":[{"type":"list","items":["Fuzzy control systems: used extensively in consumer appliances (washing machines, air conditioners, rice cookers) and industrial process control, where fuzzy if-then rules encode expert operator knowledge to control a process using linguistic terms rather than requiring a precise mathematical model of the plant.","Pattern recognition and image processing: fuzzy clustering algorithms (such as fuzzy c-means) allow data points to belong to multiple clusters with varying membership degrees, useful for image segmentation and classification tasks where boundaries between classes are inherently ambiguous.","Decision-making and multi-criteria analysis: fuzzy sets model imprecise decision criteria and preferences, allowing decision-support systems to handle vague, linguistic evaluation criteria such as 'high priority' or 'moderate risk'.","Medical diagnosis: fuzzy expert systems handle the inherent imprecision and uncertainty in medical symptoms and diagnostic criteria, supporting diagnosis where symptom severity is naturally graded rather than binary.","Fuzzy databases and information retrieval: allow queries using imprecise, linguistic search criteria (such as searching for 'affordable' housing) rather than requiring exact numerical thresholds.","Reliability and risk assessment: fuzzy set theory models component failure possibilities and risk levels expressed in imprecise linguistic terms, complementing traditional probabilistic reliability analysis particularly when historical failure data is scarce."]}]}},{"id":"soft-computing-2023-8E1850-pC-q1","number":1,"marks":15,"tags":["Part C"],"text":[{"type":"paragraph","content":"Q.1. Explain the Genetic Algorithm using a suitable example with a detailed description of the Genetic Algorithm Operators."}],"answer":{"explanation":[{"type":"heading","content":"Genetic Algorithm"},{"type":"paragraph","content":"A Genetic Algorithm (GA) is a population-based, stochastic optimization and search technique inspired by the principles of natural selection and genetics, in which a population of candidate solutions, each encoded as a chromosome (typically a binary string, though real-valued and other encodings are also used), evolves over successive generations toward increasingly better solutions to a given problem through the repeated application of selection, crossover, and mutation operators, guided by a fitness function that quantifies how good each candidate solution is at solving the target problem."},{"type":"subheading","content":"Basic Genetic Algorithm Cycle"},{"type":"list","items":["Initialization: generate an initial population of candidate solutions, typically at random, with each individual encoded as a chromosome representing a point in the search space.","Fitness evaluation: evaluate each individual's fitness by decoding its chromosome and applying the problem's objective (fitness) function, quantifying how well that individual solves the target problem.","Selection: select individuals from the current population to become parents for the next generation, with fitter individuals given a higher probability of selection (common methods include roulette wheel selection, tournament selection, and rank-based selection).","Crossover (recombination): combine pairs of selected parent chromosomes to produce offspring, exchanging genetic material between parents to create new candidate solutions that inherit characteristics from both parents.","Mutation: randomly alter a small number of genes (bits) in the offspring chromosomes with a small probability, introducing new genetic diversity into the population and helping the algorithm escape local optima.","Replacement: form the new generation's population from the offspring (and, in elitist variants, the best individuals from the previous generation), and repeat the cycle from fitness evaluation until a stopping criterion (such as a maximum number of generations, or convergence to a satisfactory fitness level) is met."]},{"type":"subheading","content":"Worked Example"},{"type":"paragraph","content":"Consider maximizing the simple function f(x) = x^2 over integers x in the range 0 to 31, encoded as a 5-bit binary chromosome. An initial population of four random chromosomes might be 01100 (x=12, f=144), 11001 (x=25, f=625), 00101 (x=5, f=25), and 10011 (x=19, f=361). Selection (say, roulette wheel selection weighted by fitness) would favor the chromosome 11001 (fitness 625) and 10011 (fitness 361) as parents more often than the lower-fitness individuals. Applying single-point crossover between selected parents 11001 and 10011 at, say, position 3, produces offspring 11011 (x=27, f=729) and 10001 (x=17, f=289), illustrating how crossover can combine high-order bits from one parent with lower-order bits from another to potentially discover an even fitter individual (729 exceeds both parents' fitness). Mutation might then flip a randomly selected bit in one offspring, for instance changing 11011 to 11111 (x=31, f=961), the maximum possible fitness for this problem, illustrating how mutation introduces genetic material not present in either parent and can occasionally produce a substantial fitness improvement, complementing crossover's recombination-based search."},{"type":"subheading","content":"Genetic Algorithm Operators in Detail"},{"type":"paragraph","content":"Selection operators determine which individuals get to reproduce, with the general principle that fitter individuals should have a higher probability of contributing offspring to the next generation, but without deterministically excluding weaker individuals entirely (to maintain genetic diversity). Roulette wheel selection assigns each individual a probability of selection proportional to its fitness relative to the total population fitness, analogous to a roulette wheel with sectors sized according to fitness. Tournament selection randomly picks a small subset of individuals and selects the fittest among them, repeated as many times as needed to fill the mating pool, offering good control over selection pressure through the tournament size parameter."},{"type":"paragraph","content":"Crossover operators recombine genetic material from two parent chromosomes to produce offspring. Single-point crossover selects a single random crossover point along the chromosome length and swaps all genes beyond that point between the two parents, as illustrated in the worked example above. Two-point crossover selects two crossover points and swaps the segment of genes between them. Uniform crossover independently decides, gene by gene, which parent contributes each specific gene to the offspring, typically with equal probability for each parent at each gene position, giving a more thorough mixing of genetic material than single or two-point crossover but potentially disrupting beneficial gene combinations (building blocks) more readily."},{"type":"paragraph","content":"Mutation operators introduce random, small perturbations to individual genes, with a typically low mutation probability (commonly a fraction of a percent per gene) chosen to introduce occasional new genetic diversity without overwhelming the more directed search provided by selection and crossover. For binary-encoded chromosomes, mutation typically flips a randomly selected bit from 0 to 1 or vice versa. Mutation serves the crucial role of preventing the population from prematurely converging to a suboptimal solution by ensuring that genetic material lost during selection (due to below-average individuals being eliminated) can still be reintroduced, and that the algorithm retains some capability to explore regions of the search space not represented in the current population's gene pool."},{"type":"paragraph","content":"Together, these three operators, selection (exploitation of currently promising solutions), crossover (recombination of promising building blocks from different individuals), and mutation (exploration of new genetic material), balance the fundamental exploration-exploitation trade-off central to all evolutionary search algorithms, allowing genetic algorithms to effectively search large, complex, and multi-modal search spaces where traditional gradient-based optimization methods would readily become trapped in local optima, making genetic algorithms particularly valuable for combinatorial optimization, engineering design optimization, and machine learning hyperparameter tuning problems where the objective function may be discontinuous, non-differentiable, or otherwise poorly suited to classical calculus-based optimization techniques."}]}},{"id":"soft-computing-2023-8E1850-pC-q2","number":2,"marks":15,"tags":["Part C"],"text":[{"type":"paragraph","content":"Q.2. Explain the Fuzzy Neural Networks and their learning-Architecture of Neuro-Fuzzy Systems, Generation of Fuzzy Rules in Neuro-Fuzzy Systems."}],"answer":{"explanation":[{"type":"heading","content":"Fuzzy Neural Networks"},{"type":"paragraph","content":"Fuzzy neural networks (neuro-fuzzy systems) combine the human-interpretable, linguistic rule-based reasoning capability of fuzzy logic with the learning and adaptation capability of artificial neural networks, addressing a key limitation of each technique when used alone: pure fuzzy systems require an expert to manually specify membership functions and rules, offering no automatic learning capability, while pure neural networks learn effectively from data but produce an opaque, uninterpretable 'black box' model whose internal weights carry no direct linguistic meaning. A neuro-fuzzy system represents a fuzzy inference system's components (membership functions, rules) using a neural-network-like layered architecture whose parameters (membership function centers and widths, rule weights) can then be tuned using standard neural network learning algorithms such as backpropagation-based gradient descent, applied to training data, giving the resulting system both the interpretability of fuzzy rules and the automatic learning capability of neural networks."},{"type":"subheading","content":"Architecture of Neuro-Fuzzy Systems (ANFIS-style)"},{"type":"paragraph","content":"A typical adaptive neuro-fuzzy inference system (ANFIS) architecture is organized into five layers, each performing a specific function in the overall fuzzy inference process, structured so that the entire fuzzy inference procedure can be represented and trained as a feed-forward neural network."},{"type":"list","items":["Layer 1 (Fuzzification layer): each node computes the membership degree of a crisp input value in a corresponding fuzzy set (linguistic term), using an adjustable membership function (commonly a bell-shaped or Gaussian function) whose parameters (center, width) are adjustable, learnable weights of the network.","Layer 2 (Rule layer): each node computes the firing strength of a specific fuzzy rule by combining (typically multiplying, implementing a fuzzy AND/T-norm operation) the relevant membership degrees from Layer 1 corresponding to that rule's antecedent conditions.","Layer 3 (Normalization layer): each node normalizes a given rule's firing strength by dividing it by the sum of all rules' firing strengths, producing a normalized firing strength that sums to one across all rules.","Layer 4 (Consequent layer): each node computes the weighted contribution of a given rule to the overall output, typically as the normalized firing strength multiplied by a linear function of the inputs (in a Sugeno-style system) whose coefficients are additional adjustable, learnable parameters.","Layer 5 (Output layer): a single node sums the weighted contributions from all rules in Layer 4 to produce the network's final, defuzzified crisp output."]},{"type":"paragraph","content":"This layered structure allows the entire fuzzy inference computation to be expressed as a sequence of differentiable operations, enabling standard gradient-based learning algorithms (such as a hybrid learning rule combining gradient descent for the antecedent membership function parameters with least-squares estimation for the consequent linear function parameters, as used in the original ANFIS formulation) to automatically tune both the shape of the fuzzy membership functions and the rule consequent parameters directly from training data, rather than requiring these to be manually specified by a human expert as in a conventional fuzzy system."},{"type":"subheading","content":"Generation of Fuzzy Rules in Neuro-Fuzzy Systems"},{"type":"paragraph","content":"Fuzzy rule generation in a neuro-fuzzy system can proceed through several distinct strategies. Grid partitioning divides each input variable's domain into a fixed number of fuzzy sets (linguistic terms), and forms one rule for every possible combination of these fuzzy sets across all input variables, giving a complete, systematic rule base but suffering from a combinatorial explosion in the number of rules as the number of input variables grows (the well-known curse of dimensionality). Clustering-based rule generation instead first applies a clustering algorithm (such as fuzzy c-means) to the training data to identify natural groupings or clusters in the input-output data, and then forms one fuzzy rule per identified cluster, with the cluster center and spread directly informing the corresponding rule's antecedent membership function parameters, producing a much more compact rule base whose size scales with the actual complexity of the data rather than combinatorially with the number of input dimensions. Learning-from-examples approaches use a training algorithm to directly and incrementally construct rules from the training data as it is presented, adding new rules when the existing rule base cannot adequately account for a new training example, and adjusting existing rule parameters otherwise, allowing the rule base to grow organically to match the complexity actually present in the training data."},{"type":"paragraph","content":"Once an initial rule base has been established through one of these methods, the neuro-fuzzy system's overall learning algorithm (the hybrid gradient-descent-and-least-squares approach used in ANFIS, or a pure backpropagation approach in simpler architectures) then fine-tunes both the membership function parameters and the rule consequent parameters to minimize the error between the network's output and the desired training output, effectively combining the initial, structurally-informed rule generation step with a subsequent, data-driven parameter refinement step to produce a final fuzzy inference system that is both interpretable (since its rules remain expressed in recognizable fuzzy if-then form) and accurate (since its parameters have been tuned directly against training data using efficient gradient-based optimization), making neuro-fuzzy systems a particularly valuable soft computing technique for applications requiring both model accuracy and human interpretability, such as medical diagnosis support systems and industrial process control, where a purely black-box neural network model would be difficult for domain experts to trust or validate."}]}},{"id":"soft-computing-2023-8E1850-pC-q3","number":3,"marks":15,"tags":["Part C"],"text":[{"type":"paragraph","content":"Q.3. Given initial solutions (X1, X2 pairs): Solution 1 (3.22, 0.40), Solution 2 (0.19, 2.28), Solution 3 (3.18, 0.33), Solution 4 (1.66, 4.59), Solution 5 (2.21, 0.86). Write the updated solutions using the PSO algorithm for one iteration. The swarm size is five solutions and the minimizing function is f(x) = x1^2 + x2^2 (assumed, as the standard textbook minimization function for this class of PSO problem when not otherwise specified). Always use a random number = 1.0 whenever required. The values of c1 and c2 are 1.0. Consider the iteration as the first iteration and the value of velocities are the same as the solutions."}],"answer":{"explanation":[{"type":"heading","content":"PSO Algorithm: One Iteration Update"},{"type":"paragraph","content":"Particle Swarm Optimization (PSO) maintains a swarm (population) of candidate solutions (particles), each with a position vector (the candidate solution itself) and a velocity vector (the direction and magnitude of the particle's movement through the search space), updated at each iteration according to the particle's own historical best position (pbest) and the entire swarm's best-known position (gbest), using the standard velocity and position update equations."},{"type":"math","content":"V_i(t+1) = w V_i(t) + c_1 r_1 (P_{best,i} - X_i(t)) + c_2 r_2 (G_{best} - X_i(t))"},{"type":"math","content":"X_i(t+1) = X_i(t) + V_i(t+1)"},{"type":"paragraph","content":"Given the five initial solutions (Solution 1: X1=3.22, X2=0.40; Solution 2: X1=0.19, X2=2.28; Solution 3: X1=3.18, X2=0.33; Solution 4: X1=1.66, X2=4.59; Solution 5: X1=2.21, X2=0.86), and using the standard minimization function f(x1,x2) = x1^2 + x2^2 (the common textbook choice for this class of introductory PSO problem, since the specific function symbol was not fully extractable from the source document image, though the calculation methodology below applies identically regardless of the exact function used, simply substituting the correct fitness values)."},{"type":"table","headers":["Solution","X1","X2","f(X) = X1^2+X2^2"],"rows":[["1","3.22","0.40","10.53"],["2","0.19","2.28","5.23"],["3","3.18","0.33","10.22"],["4","1.66","4.59","23.83"],["5","2.21","0.86","5.62"]]},{"type":"paragraph","content":"Since this is the first iteration, each particle's personal best (Pbest) is its own current position (no prior iteration exists to have found a better position yet), and the global best (Gbest) is the position of the particle with the lowest (best, for minimization) fitness value among the initial five, which is Solution 2 (X1=0.19, X2=2.28, f=5.23), the minimum fitness value in the table above."},{"type":"paragraph","content":"Since Pbest(i) = X(i) for every particle at this first iteration, the term c1*r1*(Pbest(i) - X(i)) = c1*r1*0 = 0 for every particle, meaning the velocity update reduces entirely to the global-best-attraction term for this specific first iteration: V(i+1) = V(i) + c2*r2*(Gbest - X(i)), using the given values w (assumed 1, as inertia weight was not separately specified), c1=c2=1.0, and random numbers r1=r2=1.0 as instructed."},{"type":"math","content":"V_i(t+1) = V_i(t) + 1.0 \\times 1.0 \\times (G_{best} - X_i(t))"},{"type":"paragraph","content":"Since the velocities are given as equal to the solutions themselves (V(i) = X(i) initially), the updated velocity for each particle is V(i)_new = X(i) + (Gbest - X(i)) = Gbest, for each coordinate, since the X(i) terms cancel. This gives each particle's new velocity, for both the X1 and X2 components, equal to the global best position (0.19, 2.28) for every particle in this specific first iteration, due to the given initial condition that velocity equals position."},{"type":"table","headers":["Solution","New V1","New V2"],"rows":[["1","0.19","2.28"],["2","0.19","2.28"],["3","0.19","2.28"],["4","0.19","2.28"],["5","0.19","2.28"]]},{"type":"paragraph","content":"Finally, the updated position for each particle is X(i)_new = X(i)_old + V(i)_new."},{"type":"table","headers":["Solution","New X1","New X2"],"rows":[["1","3.22+0.19=3.41","0.40+2.28=2.68"],["2","0.19+0.19=0.38","2.28+2.28=4.56"],["3","3.18+0.19=3.37","0.33+2.28=2.61"],["4","1.66+0.19=1.85","4.59+2.28=6.87"],["5","2.21+0.19=2.40","0.86+2.28=3.14"]]},{"type":"paragraph","content":"These five updated (X1, X2) position pairs represent the swarm's state after one complete PSO iteration; in the next iteration, each particle's fitness would be re-evaluated at its new position, personal bests would be updated for any particle whose new fitness improves upon its previous personal best, the global best would be updated if any particle's new position yields a fitness better than the current global best, and the velocity and position update equations would be applied again using these updated Pbest and Gbest values, repeating the entire cycle for as many iterations as needed until the swarm converges to a satisfactory minimum of the objective function or a maximum iteration count is reached."}]}},{"id":"soft-computing-2023-8E1850-pC-q4","number":4,"marks":15,"tags":["Part C"],"text":[{"type":"paragraph","content":"Q.4. (a) Explain the methods used for the decomposition of compound linguistic rules into simple canonical rules. [8]"},{"type":"paragraph","content":"(b) List the basic logic operations performed over the propositions. [7]"}],"answer":{"explanation":[{"type":"heading","content":"Decomposition of Compound Linguistic Rules"},{"type":"paragraph","content":"A compound linguistic (fuzzy) rule combines multiple antecedent conditions or multiple consequent actions using logical connectives (AND, OR) within a single if-then statement, such as 'if temperature is high AND pressure is high then valve is fully-open', and decomposition refers to the systematic process of breaking down such a compound rule into a set of simpler, canonical single-antecedent, single-consequent rules that a fuzzy inference engine can process directly."},{"type":"subheading","content":"Conjunctive (AND) Rule Decomposition"},{"type":"paragraph","content":"A rule of the form 'if x is A and y is B then z is C' is treated as a single canonical rule whose antecedent is itself a compound fuzzy set formed by the fuzzy intersection (typically the minimum, or T-norm, operation) of the membership degrees of A and B evaluated at the current input values; the rule's firing strength is then computed as min(mu_A(x), mu_B(y)), and this single firing strength drives the single consequent C. This is the most common decomposition treatment for AND-connected antecedents in standard Mamdani and Sugeno fuzzy inference systems, and requires no further decomposition into separate rules, since the AND operation is handled directly within the T-norm combination of the antecedent membership degrees."},{"type":"subheading","content":"Disjunctive (OR) Rule Decomposition"},{"type":"paragraph","content":"A rule of the form 'if x is A or y is B then z is C' can either be handled directly using a fuzzy union (typically the maximum, or T-conorm, operation) applied to the antecedent membership degrees, giving a firing strength of max(mu_A(x), mu_B(y)) driving the single consequent C, or alternatively decomposed into two entirely separate canonical rules, 'if x is A then z is C' and 'if y is B then z is C', with the outputs of these two separate rules subsequently aggregated (typically via the maximum operation) during the overall fuzzy inference aggregation step; both approaches are mathematically equivalent under the standard max-based fuzzy union and aggregation operations, and the choice between them is largely a matter of implementation convenience."},{"type":"subheading","content":"Rules with Compound Consequents"},{"type":"paragraph","content":"A rule with a compound (AND-connected) consequent, such as 'if x is A then y is B and z is C', is decomposed into two separate canonical rules sharing the same antecedent but each with a single consequent: 'if x is A then y is B' and 'if x is A then z is C', since a fuzzy inference engine's rule evaluation and aggregation mechanism is generally structured to handle one output variable (consequent) per rule, requiring rules affecting multiple distinct output variables to be split into separate single-consequent rules, each processed independently for its own respective output variable."},{"type":"paragraph","content":"Nested and hierarchical compound rules, involving multiple levels of AND/OR combination among several antecedent variables, are decomposed by first applying the appropriate T-norm (for AND) or T-conorm (for OR) operations according to the rule's logical structure (respecting parentheses and operator precedence exactly as in ordinary Boolean logic expressions) to compute a single overall antecedent firing strength, and this firing strength then drives the rule's consequent exactly as in the simpler cases discussed above, meaning arbitrarily complex compound linguistic rules can, in principle, always be reduced to this same canonical single-firing-strength, single-consequent form through repeated application of the T-norm and T-conorm combination rules."},{"type":"heading","content":"Basic Logic Operations Over Propositions"},{"type":"list","items":["Negation (NOT): for a fuzzy proposition 'x is A', the negation 'x is NOT A' has membership function mu_notA(x) = 1 - mu_A(x), the standard fuzzy complement operation.","Conjunction (AND): the truth value of 'P AND Q' for fuzzy propositions P and Q is computed using a T-norm operation, most commonly the minimum, T(a,b) = min(a,b), though other T-norms such as the algebraic product T(a,b) = a*b are also used in specific applications.","Disjunction (OR): the truth value of 'P OR Q' is computed using a T-conorm (S-norm) operation, most commonly the maximum, S(a,b) = max(a,b), though other T-conorms such as the algebraic sum S(a,b) = a+b-ab are also used.","Implication (IF-THEN): the truth value of the fuzzy implication 'if P then Q' is computed using one of several fuzzy implication operators, such as the Mamdani (minimum) implication, min(mu_P, mu_Q), or the Zadeh implication, max(1-mu_P, min(mu_P, mu_Q)), or the Lukasiewicz implication, min(1, 1-mu_P+mu_Q), with the choice of implication operator affecting the specific behavior of the resulting fuzzy inference system.","Equivalence (IFF): the truth value of 'P if and only if Q' is typically computed as the fuzzy conjunction of the two implications in each direction, min(Implication(P,Q), Implication(Q,P)), representing mutual, bidirectional implication between the two fuzzy propositions."]},{"type":"paragraph","content":"These basic fuzzy logic operations, negation, conjunction, disjunction, implication, and equivalence, together form the complete operational toolkit needed to evaluate arbitrarily complex compound fuzzy propositions and rules, extending classical Boolean propositional logic's operations to the continuous-valued (0 to 1) truth domain of fuzzy logic, and forming the mathematical foundation on which the rule decomposition techniques discussed in the first part of this question, and the broader fuzzy inference process discussed throughout this examination, are built."}]}},{"id":"soft-computing-2023-8E1850-pC-q5","number":5,"marks":15,"tags":["Part C"],"text":[{"type":"paragraph","content":"Q.5. The following data was determined by the pairwise comparison of work preferences of 100 people: When it was compared with software (S), 40 persons polled preferred hardware (H), 38 of them preferred teaching (T), 48 of them preferred business (B) and 30 preferred textiles (TX). In comparison with hardware (H), the preferences were 60 for S, 52 for T, 53 for B and 35 for TX. When compared with teaching, the preferences were 62 for S, 48 for H, 65 for B and 56 for TX. In comparison with business, the preferences were 52 for S, 47 for H, 35 for T and 60 for TX. When compared with textile, the preferences were 70 for S, 65 for H, 44 for T and 40 for B. Using the rank ordering, plot the membership function for the 'most preferred work.'"}],"answer":{"explanation":[{"type":"heading","content":"Rank Ordering Method for Membership Function Determination"},{"type":"paragraph","content":"The rank ordering method determines fuzzy membership values from pairwise comparison (preference) data collected from a group of respondents, converting subjective, pairwise preference judgments into a single, consistent membership function over the set of alternatives being compared, based on the principle that an alternative preferred more often (and more strongly) across all pairwise comparisons should receive a correspondingly higher membership degree in the fuzzy set representing the target linguistic concept, here 'most preferred work.'"},{"type":"subheading","content":"Constructing the Pairwise Preference Matrix"},{"type":"paragraph","content":"The given data specifies, for each pair of alternatives, how many of the 100 polled people preferred one over the other. Organizing this into a complete pairwise preference matrix (where each entry represents the number of people preferring the row alternative over the column alternative, out of the total 100 respondents), using the given data: comparing S against others, 40 preferred H over S, 38 preferred T over S, 48 preferred B over S, 30 preferred TX over S, meaning S's own preference count against each is the complement out of 100 in each specific pairwise poll (60 preferred S over H, 62 preferred S over T, 52 preferred S over B, 70 preferred S over TX, using the complementary data given directly for the H, T, B, and TX rows)."},{"type":"table","headers":["vs S","vs H","vs T","vs B","vs TX","Row Sum"],"rows":[["S: -","40","38","48","30","156"],["H: 60","-","52","53","35","200"],["T: 62","48","-","65","56","231"],["B: 52","47","35","-","60","194"],["TX: 70","65","44","40","-","219"]]},{"type":"paragraph","content":"Each row sum is computed by adding all the entries in that row (the number of times, across all four pairwise comparisons involving that alternative, that this alternative was preferred): Row sum for S = 40+38+48+30 = 156; Row sum for H = 60+52+53+35 = 200; Row sum for T = 62+48+65+56 = 231; Row sum for B = 52+47+35+60 = 194; Row sum for TX = 70+65+44+40 = 219."},{"type":"math","content":"\\text{Total of all row sums} = 156+200+231+194+219 = 1000"},{"type":"subheading","content":"Computing Membership Values"},{"type":"paragraph","content":"The membership degree of each alternative in the fuzzy set 'most preferred work' is computed by normalizing each alternative's row sum by the grand total of all row sums (1000 in this case, since each of the 5 alternatives contributes 4 pairwise comparisons of up to 100 each, giving a maximum possible total of 5*4*100/2 = 1000 when counted this way, consistent with the computed total above): mu(alternative) = row_sum(alternative) / total_of_all_row_sums."},{"type":"math","content":"\\mu(S) = 156/1000 = 0.156, \\quad \\mu(H) = 200/1000 = 0.200"},{"type":"math","content":"\\mu(T) = 231/1000 = 0.231, \\quad \\mu(B) = 194/1000 = 0.194, \\quad \\mu(TX) = 219/1000 = 0.219"},{"type":"table","headers":["Alternative","Row Sum","Membership mu"],"rows":[["Software (S)","156","0.156"],["Hardware (H)","200","0.200"],["Teaching (T)","231","0.231"],["Business (B)","194","0.194"],["Textiles (TX)","219","0.219"]]},{"type":"subheading","content":"Membership Function Plot"},{"type":"svg","alt":"Bar chart showing the rank-ordering-derived membership function values for each work preference alternative, with Teaching having the highest membership value","content":""},{"type":"paragraph","content":"The resulting membership function, plotted as a bar chart across the five alternatives, shows Teaching (T) with the highest membership value (0.231) in the fuzzy set 'most preferred work,' followed closely by Textiles (0.219), Hardware (0.200), Business (0.194), and finally Software (0.156) with the lowest membership degree among the five alternatives; this ranking reflects the aggregate outcome of all pairwise preference comparisons collected from the 100 respondents, converted through the rank ordering method into a properly normalized fuzzy membership function whose values sum to 1.0 across all five alternatives (0.156+0.200+0.231+0.194+0.219 = 1.000), consistent with the requirement that a properly normalized membership distribution over a complete, mutually exclusive set of alternatives sum to unity. This rank-ordering-derived membership function can then be used directly as a fuzzy set input to further fuzzy decision-making or multi-criteria analysis procedures, wherever a numerical representation of aggregate group preference (as opposed to a simple majority-vote ranking, which would discard the relative strength of preference information captured by the raw pairwise comparison counts) is required."}]}}]},{"id":"soft-computing-2022-8E1850","title":"Soft Computing (June 2022)","subjectSlug":"soft-computing","subjectName":"Soft Computing","universitySlug":"rtu","branchSlug":"ee-ec-eee","year":4,"semester":8,"paperCode":"8E1850","examYear":2022,"questions":[{"id":"soft-computing-2022-8E1850-pA-q1","number":1,"marks":2,"tags":["Part A"],"text":[{"type":"paragraph","content":"1. Define the aims of Soft Computing."}]},{"id":"soft-computing-2022-8E1850-pA-q2","number":2,"marks":2,"tags":["Part A"],"text":[{"type":"paragraph","content":"2. Give Applications of Fuzzy Sets."}]},{"id":"soft-computing-2022-8E1850-pA-q3","number":3,"marks":2,"tags":["Part A"],"text":[{"type":"paragraph","content":"3. Define Union Fuzzy Operation."}]},{"id":"soft-computing-2022-8E1850-pA-q4","number":4,"marks":2,"tags":["Part A"],"text":[{"type":"paragraph","content":"4. Define Artificial Neural Network."}]},{"id":"soft-computing-2022-8E1850-pA-q5","number":5,"marks":2,"tags":["Part A"],"text":[{"type":"paragraph","content":"5. Define Search Space and Optimization."}]},{"id":"soft-computing-2022-8E1850-pA-q6","number":6,"marks":2,"tags":["Part A"],"text":[{"type":"paragraph","content":"6. Define Fuzzy logic."}]},{"id":"soft-computing-2022-8E1850-pA-q7","number":7,"marks":2,"tags":["Part A"],"text":[{"type":"paragraph","content":"7. Define Genetic Operator - Crossover and Mutation."}]},{"id":"soft-computing-2022-8E1850-pA-q8","number":8,"marks":2,"tags":["Part A"],"text":[{"type":"paragraph","content":"8. Define Fuzzy Neural Networks."}]},{"id":"soft-computing-2022-8E1850-pA-q9","number":9,"marks":2,"tags":["Part A"],"text":[{"type":"paragraph","content":"9. Define Genetic Algorithms."}]},{"id":"soft-computing-2022-8E1850-pA-q10","number":10,"marks":2,"tags":["Part A"],"text":[{"type":"paragraph","content":"10. Define Neuro-Fuzzy Identification."}]},{"id":"soft-computing-2022-8E1850-pB-q1","number":1,"marks":8,"tags":["Part B"],"text":[{"type":"paragraph","content":"1. Describe Fuzzy Operations discussing following with examples: (i) Inclusion (ii) Equality (iii) Complement (iv) Intersection."}]},{"id":"soft-computing-2022-8E1850-pB-q2","number":2,"marks":8,"tags":["Part B"],"text":[{"type":"paragraph","content":"2. Discuss and compare Supervised, Unsupervised and Reinforcement Learning in connection with Artificial Neural Network (ANN)."}]},{"id":"soft-computing-2022-8E1850-pB-q3","number":3,"marks":8,"tags":["Part B"],"text":[{"type":"paragraph","content":"3. Discuss Radial Basis Neural Network along with its advantages."}]},{"id":"soft-computing-2022-8E1850-pB-q4","number":4,"marks":8,"tags":["Part B"],"text":[{"type":"paragraph","content":"4. Discuss Adaptive Resonance Theory (ART) along with its advantages."}]},{"id":"soft-computing-2022-8E1850-pB-q5","number":5,"marks":8,"tags":["Part B"],"text":[{"type":"paragraph","content":"5. Explain and contrast Conventional and Genetic Search Algorithms."}]},{"id":"soft-computing-2022-8E1850-pB-q6","number":6,"marks":8,"tags":["Part B"],"text":[{"type":"paragraph","content":"6. Describe Genetic Algorithm based Optimization. Also explain Genetic Algorithm with mutation."}]},{"id":"soft-computing-2022-8E1850-pB-q7","number":7,"marks":8,"tags":["Part B"],"text":[{"type":"paragraph","content":"7. Discuss learning and architecture of Neuro-Fuzzy Systems."}]},{"id":"soft-computing-2022-8E1850-pC-q1","number":1,"marks":15,"tags":["Part C"],"text":[{"type":"paragraph","content":"1. Explain Fuzzy Sets along with its properties. Also discuss Fuzzy Relations."}]},{"id":"soft-computing-2022-8E1850-pC-q2","number":2,"marks":15,"tags":["Part C"],"text":[{"type":"paragraph","content":"2. Explain the following architectures of Artificial Neural Networks (ANN) along with their advantages and disadvantages: (i) Single Layer Feed Forward Neural Networks (ii) Multi-Layer Feed Forward Neural Networks (iii) Recurrent Neural Networks."}]},{"id":"soft-computing-2022-8E1850-pC-q3","number":3,"marks":15,"tags":["Part C"],"text":[{"type":"paragraph","content":"3. Explain Particle Swarm Optimization (PSO) and its application in engineering."}]},{"id":"soft-computing-2022-8E1850-pC-q4","number":4,"marks":15,"tags":["Part C"],"text":[{"type":"paragraph","content":"4. Write short note on: (i) Fuzzification in Neuro-Fuzzy Systems (ii) Defuzzification Neuro-Fuzzy Systems."}]},{"id":"soft-computing-2022-8E1850-pC-q5","number":5,"marks":15,"tags":["Part C"],"text":[{"type":"paragraph","content":"5. Write short note on any two: (i) Neural Network Toolbox in MATLAB (ii) Fuzzy Logic Toolbox in MATLAB (iii) Genetic Algorithm and Directed Search Toolbox in MATLAB."}]}]}],"loading":false,"error":null}}}

Q3Soft Computing

Question

Answer