IT-701 Soft Computing - Dec 2024 Question Paper

Under each question, How to answer lists the points an answer should cover, outlined by CampusPrep from the syllabus. Check the details against your textbook.

1. 7 Marks
   aState any two features of Genetic algorithm. Also discuss Parallel Genetic algorithm.
   How to answer

   • Define genetic algorithm and list any two features (population-based, stochastic operators, fitness-based selection)
   • Explain selection, crossover, mutation briefly
   • Define parallel genetic algorithm concept and need for speedup
   • Discuss types / implementation models like master-slave, island, fine-grained
2. 7 Marks
   bWhat is a Convolution Neural Network (CNN) and how does it work?
   How to answer

   • Define CNN and its purpose for grid-like data / images
   • Describe layers: convolution, activation, pooling, fully connected
   • Explain weight sharing and feature map learning
   • Explain working with forward pass and training example
3. 7 Marks
   aWhat is the Softmax function and why is it needed?
   How to answer

   • Define softmax function with formula $\sigma(z_i)=e^{z_i}/\sum_j e^{z_j}$
   • Explain conversion of logits to probability distribution
   • Explain need: multi-class classification, differentiability, cross-entropy compatibility
4. 7 Marks
   bExplain the structure of a neuron and its various components. Also discuss the Hebb Rule in the context of neural networks.
   How to answer

   • Describe biological/artificial neuron components: inputs, weights, summation, activation, output
   • Explain function of each component with block diagram
   • State Hebb rule: $\Delta w_{ij} = \eta x_i y_j$
   • Discuss Hebbian learning significance and limitations
5. 7 Marks
   aState the difference between supervised learning and unsupervised learning.
   How to answer

   • Define supervised learning with labelled data
   • Define unsupervised learning with unlabelled data
   • Compare on basis of feedback, goal, algorithms, examples
   • Give examples: classification/regression vs clustering/association
6. 7 Marks
   bExplain the architecture and algorithm of counter propagation network with diagram.
   How to answer

   • Describe counterpropagation network: Kohonen + Grossberg layers
   • Explain forward-only and training algorithm steps
   • Draw neat labelled architecture diagram
   • Mention applications and advantages
7. 7 Marks
   aDiscuss the application of neural network in forecasting and data compression.
   How to answer

   • Briefly define neural network for prediction/compression
   • Explain forecasting use: time-series, training, generalization
   • Explain data compression use: bottleneck/autoencoder architecture
   • Give examples of applications
8. 7 Marks
   bDiscuss Ant colony optimization method.
   How to answer

   • Introduce ant colony optimization as swarm intelligence metaheuristic
   • Explain pheromone trail, heuristic, probabilistic transition
   • Outline algorithm steps: initialization, tour construction, pheromone update
   • Mention applications e.g. TSP, routing
9. 7 Marks
   aConsider two fuzzy sets. $$A_1 = 0.2/X_1 + 0.9/X_2$$ $$A_2 = 0.3/Y_1 + 0.5/Y_2 + 1/Y_3$$ Determine $A_1 - A_2$ OR $A_1 + A_2$
   How to answer

   • Recall fuzzy Cartesian product / composition for $A_1 \times A_2$
   • Compute memberships using min for $A_1 - A_2$ (difference) interpretation
   • Compute memberships using max for $A_1 + A_2$ (union) interpretation
   • Present resulting fuzzy relations on $X \times Y$
10. 7 Marks
    bExplain error back propagation training algorithm with the help of flow chart.
    How to answer

    • State backpropagation concept: gradient descent + chain rule
    • List training steps: forward pass, error calculation, backward weight update
    • Draw flowchart showing initialization to convergence
    • Mention stopping criteria and learning rate role
11. 7 Marks
    aExplain single layer neural network architecture using perceptron model with suitable activation function.
    How to answer

    • Draw single-layer perceptron architecture: inputs, weights, bias, summation
    • Define net input and activation function e.g. step, sigmoid
    • Explain training / weight update rule
    • Mention linearly separable limitation
12. 7 Marks
    bWrite short notes on :

    1. i)Lambda-cuts for fuzzy sets
    2. ii)Neuro hybrid system
    How to answer

    • Define lambda-cut $A_\lambda = \{x | \mu_A(x) \ge \lambda\}$ with example
    • Explain properties and use in decomposition
    • Define neuro hybrid / neuro-fuzzy system concept
    • Mention architecture and advantages of hybrid system
13. 7 Marks
    aClassify the various types of soft computing techniques and mention some application areas for NN.
    How to answer

    • Classify soft computing: fuzzy logic, neural networks, evolutionary, probabilistic
    • Briefly describe each technique
    • List NN application areas: pattern recognition, control, forecasting etc.
    • Contrast hard vs soft computing if needed
14. 7 Marks
    bWrite brief notes on Kohonen's self organizing networks.
    How to answer

    • Define Kohonen self-organizing map as unsupervised competitive learning
    • Explain architecture: input to 2D lattice, winner-takes-all
    • Outline training steps: competition, cooperation, adaptation
    • Mention applications in clustering/visualization
15. 7 Marks
    aMention the role of fitness function in genetic algorithm and what are the requirements of genetic algorithm?
    How to answer

    • Define fitness function as measure of solution quality
    • Explain role in selection, guiding evolution, convergence
    • List requirements of GA: encoding, population, operators, parameters
    • Mention termination criteria
16. 7 Marks
    bDistinguish between recurrent and non-recurrent networks.
    How to answer

    • Define recurrent networks with feedback/cycles
    • Define non-recurrent/feedforward networks without cycles
    • Compare on architecture, memory, training, dynamics
    • Give examples: Hopfield/LSTM vs perceptron/CNN