About this previous year paper
Course: B.Tech (All Branches – Common Elective)
Subject Code: EOE041
Subject Title: Introduction to Soft Computing (Neural Network, Fuzzy Logic & Genetic Algorithm)
Exam Type: Theory
Duration: 3 Hours
Maximum Marks: 100
SECTION – A (10 × 2 = 20 Marks)
Short and conceptual questions from Neural Networks, Fuzzy Systems, and Genetic Algorithms.
| No. | Topic | Concept Summary |
|---|---|---|
| (a) | AI in Neural Networks | Artificial Intelligence forms the foundation for neural networks, allowing systems to learn and adapt based on data. Neural networks are AI models that simulate human brain behavior. |
| (b) | Applications of Neural Networks | Used in pattern recognition, medical diagnosis, speech processing, autonomous systems, and financial forecasting. |
| (c) | Reinforcement Learning | A type of learning based on feedback and reward; the system learns optimal behavior through trial and error. |
| (d) | Convergence of GA | Occurs when the population reaches an optimal or near-optimal solution after several generations. |
| (e) | Fuzzy Quantifiers | Linguistic terms like “most”, “few”, “many” used in fuzzy logic to quantify approximate reasoning. |
| (f) | Fuzzy Inference | Process of deriving fuzzy output from fuzzy inputs using if–then rules and logical operations. |
| (g) | Mutation (GA) | Random alteration of one or more genes in a chromosome to maintain genetic diversity. |
| (h) | Hebb Rule Example | For storing vector [1 1 1 -1] → weight matrix W=XTXW = X^T XW=XTX, excluding diagonal terms. |
| (i) | FLC (Fuzzy Logic Controller) | A controller using fuzzy logic for decision-making under uncertainty; used in control systems like ACs or car braking. |
| (j) | Benefit of GA | Provides global search capability, avoids local minima, works well with complex or non-differentiable functions. |
SECTION – B (5 × 10 = 50 Marks)
Descriptive and numerical questions testing in-depth conceptual understanding.
Key Topics:
(a) Artificial Neural Network (ANN):
Defined as a massively parallel system of interconnected neurons capable of learning from data.
Characteristics: Learning ability, adaptivity, non-linearity, fault tolerance, generalization.
(b) Backpropagation Neural Network (BPN):
A supervised learning algorithm minimizing mean square error via gradient descent.
Training factors: Learning rate, momentum, number of hidden layers, initialization weights, and activation functions.
(c) Fuzzy Set Operations:
Union: μA∪B(x)=max[μA(x),μB(x)]μ_{A∪B}(x) = max[μ_A(x), μ_B(x)]μA∪B(x)=max[μA(x),μB(x)]
Intersection: μA∩B(x)=min[μA(x),μB(x)]μ_{A∩B}(x) = min[μ_A(x), μ_B(x)]μA∩B(x)=min[μA(x),μB(x)]
Complement: μ¬A(x)=1−μA(x)μ_{¬A}(x) = 1 - μ_A(x)μ¬A(x)=1−μA(x)
Example: Let μA(x)=0.6,μB(x)=0.8μ_A(x)=0.6, μ_B(x)=0.8μA(x)=0.6,μB(x)=0.8 ⇒ A∪B=0.8,A∩B=0.6A∪B=0.8, A∩B=0.6A∪B=0.8,A∩B=0.6.
(d) BPN Parameter Selection:
Learning rate controls speed of convergence.
Momentum reduces oscillation.
More hidden neurons increase accuracy but may cause overfitting.
(e) Genetic Algorithm (GA):
A stochastic optimization algorithm inspired by natural selection.
Flowchart:
Initialize population
Evaluate fitness
Selection
Crossover
Mutation
Termination condition
(f) Roulette Wheel Selection (Fitness vs Random Based):
Fitness-based: Probability ∝ fitness value.
Random-based: Probability uniformly distributed.
Example: Higher fitness → larger “wheel segment”.
(g) Perceptron Classification:
Given:
Class 1 → (1,1,1,1), (-1,1,-1,-1)
Class -1 → (1,1,1,-1), (1,-1,-1,1)
Learning rate (η) = 1, initial weights = 0
Apply perceptron learning rule:
- wnew=wold+η(t−y)xw_{new} = w_{old} + η(t - y)xwnew=wold+η(t−y)x
to calculate final weights for classification.
(h) Predicate Logic Inference Example:
Statements:
All men are mortal.
Socrates is a man.
Therefore, Socrates is mortal.
→ Proven using modus ponens in first-order logic.
SECTION – C (2 × 15 = 30 Marks)
Analytical and applied questions from all three domains.
Q3. Neural Network Architectures
(i) Rosenblatt’s Perceptron:
Linear classifier; single-layer network.
Learning rule updates weights to minimize classification error.
(ii) McCulloch–Pitts Model:
Simplest neuron model using threshold logic.
Performs basic logical operations like AND, OR, NOT.
Q4. Fuzzy Logic Application – Greg Viot’s Fuzzy Cruise Controller
Used in automobiles for automatic speed regulation.
Components:
Fuzzification: Converts inputs (speed, acceleration) to fuzzy sets.
Inference Engine: Applies fuzzy rules (e.g., “If speed < desired then accelerate”).
Defuzzification: Converts fuzzy output to crisp control action.
Q5. Genetic Algorithm for Optimization
Problem:
Minimize (x−2.5)2+(y−5)2\text{Minimize } (x - 2.5)^2 + (y - 5)^2 Minimize (x−2.5)2+(y−5)2
Subject to:
5.5x+2y2−18≤0,0≤x, y≥55.5x + 2y^2 - 18 \le 0,\quad 0 \le x,\ y \ge 55.5x+2y2−18≤0,0≤x, y≥5
Steps:
Encode x,yx, yx,y as binary strings.
Evaluate fitness f(x,y)f(x,y)f(x,y).
Apply selection, crossover, and mutation.
Use constraints as penalty terms in fitness evaluation.
Continue until population converges to optimal (x=2.5,y=5)(x=2.5, y=5)(x=2.5,y=5).
Summary
This EOE041 – Introduction to Soft Computing exam assesses understanding of hybrid intelligent systems combining:
| Domain | Focus | Example |
|---|---|---|
| Neural Networks | Learning & pattern recognition | Backpropagation, Perceptron |
| Fuzzy Logic | Handling uncertainty | Fuzzy Controllers, Fuzzy Inference |
| Genetic Algorithms | Optimization & search | Adaptive evolution algorithms |