(SEM V) THEORY EXAMINATION 2024-25 APPLICATION OF SOFT COMPUTING

Subject Code: BCS056
Maximum Marks: 70
Time: 3 Hours
Paper ID: 310311

Question Paper Overview

SECTION A (2 × 7 = 14 Marks)

(Short conceptual questions — direct definitions and brief explanations)

a. Define single-layer and multilayer feed-forward networks.
b. What is recurrent neural network architecture?
c. State two factors that affect backpropagation training.
d. Mention one application of backpropagation networks.
e. Write two examples of fuzzy-to-crisp conversion.
f. Explain the concept of fuzzy implications.
g. List the steps involved in the generational cycle of a Genetic Algorithm (GA).

SECTION B (Attempt any three × 7 = 21 Marks)

a. Explain the concept of convergence in the context of the perceptron learning rule.
b. Describe the effect of the learning rate coefficient on training in backpropagation networks.
c. Describe the basic concepts of fuzzy logic with examples.
d. Discuss the process of fuzzification and defuzzification with examples.
e. Describe the role of genetic operators (selection, crossover, mutation) in the GA process.

SECTION C (Attempt one part from each question × 7 = 35 Marks)

Q3

(a) Describe in detail the various learning techniques used in neural networks.
OR
(b) Explain auto-associative and hetero-associative memory with examples.

Q4

(a) Compare and contrast single-layer and multilayer perceptron models.
OR
(b) Explain how backpropagation methods solve the limitations of the perceptron model.

Q5

(a) Explain the role of fuzzy relations in decision-making with examples.
OR
(b) Provide a detailed analysis of fuzzy logic’s importance in Artificial Intelligence.

Q6

(a) Describe in detail the algorithms involved in fuzzification and defuzzification.
OR
(b) Analyze the role of fuzzy controllers in industrial automation with examples.

Q7

(a) Provide a detailed explanation of industrial applications of Genetic Algorithms (GAs).
OR
(b) Compare Genetic Algorithms with traditional optimization techniques.

Key Topics for Revision

1. Feed-Forward Networks

Single-layer: Input → Output (no hidden layers).

Example: Perceptron.

Multilayer: Includes hidden layers for non-linear classification.

Example: Multilayer Perceptron (MLP) using backpropagation.

2. Recurrent Neural Networks (RNN)

Contains feedback loops → allows memory of previous outputs.

Used in time series prediction, speech recognition, and sequence modeling.

3. Backpropagation Training Factors

Learning rate (η): Determines step size in weight update.

Momentum term (α): Controls oscillations and accelerates convergence.

4. Fuzzy Logic Basics

Deals with reasoning that is approximate, not fixed (0 or 1).

Fuzzification: Converts crisp input to fuzzy sets.

Defuzzification: Converts fuzzy results back to crisp output.

Example: Temperature control – “Hot”, “Warm”, “Cool”.

5. Fuzzy Implications

Express relationships between fuzzy propositions.

Example:

“If temperature is high, then fan speed is high.”

Represented mathematically using Min–Max composition or Mamdani implication.

6. Fuzzy to Crisp Conversion

Common methods:

Centroid method (Center of Gravity).

Mean of Maximum (MoM).

7. Genetic Algorithm (GA)

Steps in Generational Cycle:

Initialize population           Evaluate fitness

Select parents                    Apply crossover

Apply mutation                  Form new population

8. Learning Techniques in Neural Networks

9. Perceptron vs Multilayer Perceptron

10. Fuzzy Relations in Decision-Making

Fuzzy relations model uncertainty and vagueness in decision variables.

Used in multi-criteria decision-making (MCDM), risk analysis, industrial automation.

11. Fuzzy Controllers

Components:

Fuzzifier                                        Rule base

Inference engine                          Defuzzifier

Applications:

Washing machines (load detection)      Air conditioners (temperature control)

Process automation

12. Fuzzification & Defuzzification Algorithms

Fuzzification:

Converts input x → membership value μ(x).

Uses membership functions: triangular, trapezoidal, Gaussian.

Defuzzification:

Converts fuzzy output → crisp value.

Methods: Centroid, Weighted Average, Max–Min.

13. Genetic Algorithms (GAs)

14. GA vs Traditional Optimization

15. Industrial Applications of GA

Optimization: Job scheduling, logistics, resource allocation.

Design: Neural network parameter tuning, antenna design.

Control Systems: Adaptive control, PID tuning.

Machine Learning: Feature selection and hyperparameter optimization.