Neural Networks Model Question

Section A

Attempt any two questions. (2 × 10 = 20)

1. Differentiate between small scale and large scale learning problems. How can heuristic be implemented for making back propagation algorithm perform better? (4+6)

2. State Cover’s theorem on separability of problems. Explain hybrid learning technique for RBF network. (3+7)

3. What is vanishing gradient problem in recurrent networks? How can it be solved? Explain with necessary equations. (4+6)

Section B

Attempt any eight questions. (8 × 5 = 40)

4. Define neural network. Briefly explain the working mechanism of biological neural with its related functional units (1+4)

5. What is a perceptron? Explain batch perceptron algorithm. (1+4)

6. Highlight on minimum-description length principle. Explain instrumental-variables method. (1+4)

7. Explain LMS algorithm. How does it differ from wiener filter.(4+1)

8. What are the properties of feature map? Explain Kernel Self-Organizing map.(1+4)

9. What is universal approximation theorem? How can real-time recurrent learning be achieved.(1+4)

10. Explain hybrid learning concept in RBF networks (5)

11. Differentiate between batch learning and on-line learning. How is learning rate controlled by using optimal annealing? Explain the concept of network pruning. (1+2+2)

12. Write short notes on: (2 × 2.5 = 5)

a. Convolutional networks

b. cross validation