Neural Networks 2078

Attempt any two questions.

1. Differentiate between Quasi-Newton method and nonlinear conjugate gradient algorithm for supervised training of multiplayer perception and explain conjugate gradient algorithm. (3+7)

2. How XOR (Exclusive OR) problem can be solved by Radial-Basis Function networks. Draw necessary figures and perform required calculations to complete the specificatioin of the RBF network. (10)

3. Explain real time recurrent learning algorithm for training a recurrent network with practical example. (10)

section B

Attempt any eight questions.

4. What are the three basic rules that depicts the flow of signal in a neural network viewed as directed graph? (5)

6. What are the assumptions that need to be considered while estimating parameter in a Gaussian environment? (5)

7. Draw the block diagram of signal-flow graph representation of the LMS algorithm and express the evolution of the weight vector. (5)

8. Describe four heuristics that provide guidelines for acclerating the convergence of back-propagation learning through learning rate adaption. (5)

9. Differentiate between RBF network and MLP network. (5)

10. Differentiate between willshaw-von der Malshurg's model and Kohonen model. (5)

11. Explain Theroem 1 with respect to the computational power of a recurrent network. (5)

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

a. Wiener filter

b. Supervised learning