B.Tech (Sem V) – Theory Notes & Answers
SECTION A – Short Answer Type
a. Soft Computing
Soft Computing is a collection of techniques such as Fuzzy Logic, Neural Networks, and Genetic Algorithms that deal with uncertainty, imprecision, and approximation.
Unlike conventional computing, it tolerates errors and provides approximate but practical solutions.
b. Auto-associative vs Hetero-associative Memory
Auto-associative memory recalls the same pattern that is given as input, while hetero-associative memory recalls a different output pattern corresponding to the input.
c. Applications of Genetic Algorithm (GA)
Genetic Algorithms are used in optimization problems, scheduling, machine learning, pattern recognition, robotics, and engineering design.
d. Self-Organizing Map (SOM)
A SOM is an unsupervised neural network that maps high-dimensional input data into a lower-dimensional space while preserving topological relationships.
e. Membership Function
A membership function defines the degree to which an element belongs to a fuzzy set. Its value ranges from 0 to 1.
f. Algebraic Sum of Fuzzy Sets
Algebraic sum is calculated as:
μ(A ⊕ B) = μA + μB − (μA × μB)
g. Basic Components of ANN
Neurons, weighted connections, activation function, bias, and learning algorithm.
h. Threshold Logic Unit (TLU)
A TLU is a neuron model that produces output based on whether the weighted sum of inputs crosses a threshold value.
i. Bias Function in Neural Network
Bias helps shift the activation function and allows better learning even when input values are zero.
j. Adaptive Learning
Adaptive learning adjusts weights dynamically based on error feedback to improve performance over time.
SECTION B – Descriptive Answers
a. Crossover in Genetic Algorithm
Crossover is a genetic operator used to combine genetic material from two parent chromosomes to create offspring.
Common types include single-point crossover, two-point crossover, and uniform crossover. It helps explore new solution spaces.
b. Weight Adjustment in Backpropagation
Weights are adjusted using gradient descent to minimize error. The error is propagated backward from output layer to input layer, and weights are updated proportionally to the error.
c. Defuzzification
Defuzzification converts fuzzy output into crisp value.
Common methods are Centroid method, Mean of Maximum, and Weighted Average method.
d. Multilayer Perceptron (MLP)
MLP consists of input, hidden, and output layers. Unlike single-layer perceptron, it can solve non-linearly separable problems.
e. Neuro-Fuzzy System
It combines neural networks and fuzzy logic. Neural networks learn rules automatically, while fuzzy logic handles uncertainty using linguistic rules.
SECTION C – Long Answer Type
Backpropagation Algorithm
Backpropagation works by computing output error, propagating it backward, adjusting weights, and repeating until minimum error is achieved.
Supervised vs Unsupervised Learning
Supervised learning uses labeled data (e.g., classification), while unsupervised learning works with unlabeled data (e.g., clustering).
XOR Problem and Limitation of Perceptron
Single-layer perceptron cannot solve XOR because XOR is not linearly separable. Multilayer networks overcome this limitation.
Kohonen Self-Organizing Network
It is an unsupervised learning network that clusters data based on similarity using competitive learning.
Fuzzy Inference System (FIS)
FIS consists of fuzzification, rule base, inference engine, and defuzzification. It converts crisp inputs into fuzzy outputs and back to crisp values.
Genetic Algorithm Working
GA follows steps: initialization, selection, crossover, mutation, and evaluation. It mimics natural evolution to find optimal solutions.
Fuzzy Sets and Properties
Fuzzy sets allow partial membership. Properties include support, core, normality, convexity, and α-cuts.
