Rr420507 Neural Networks

Set No. 1

Code No: RR420507

IV B.Tech II Semester Regular Examinations, Apr/May 2007 NEURAL NETWORKS ( Common to Computer Science & Engineering and Electronics & Computer Engineering) Time: 3 hours Max Marks: 80 Answer any FIVE Questions All Questions carry equal marks ⋆⋆⋆⋆⋆ 1. (a) Give a brief description of neural networks as optimizing networks. (b) Explain the use of ANNs for clustering and feature detection.

[8] [4+4]

2. Briefly discuss about linear separability and the solution for EX-OR problem.Also suggest a network that can solve EX-OR problem. [4+6+6] 3. Explain about the generalized delta- rule and derive the weight updatation for a multi layer feed forward neural network. [8+8] 4. Construct an energy function for a discrete Hopfield neural network of size N×N neurons. Show that the energy function decreases every time the neuron output is changed. [8+8] 5. Discuss how the “Winner-Take-All” in the Kohonen’s layer is implemented and explain the architecture, Also explain the training algorithm. [16] 6. Derive expressions for the weight updation involved in counter propagation. [16] 7. Give a detailed note on the following: (a) ART1 data structures.

[8]

(b) ART2 simulation.

[8]

8. Describe how a neural network may be trained for a pattern recognition task. Illustrate with an example [16] ⋆⋆⋆⋆⋆

1 of 1

Set No. 2

Code No: RR420507

IV B.Tech II Semester Regular Examinations, Apr/May 2007 NEURAL NETWORKS ( Common to Computer Science & Engineering and Electronics & Computer Engineering) Time: 3 hours Max Marks: 80 Answer any FIVE Questions All Questions carry equal marks ⋆⋆⋆⋆⋆ 1. (a) What is the Hebbian-learning rule for training neural networks? Explain with the help of an illustration. [4+4] (b) What is the delta learning rule in neural networks? Explain with the help of an Illustration. [4+4] 2. State and prove the perceptron convergence theorem.

[2+14]

3. Explain about the generalized delta- rule and derive the weight updatation for a multi layer feed forward neural network. [8+8] 4. The truncated energy function, E(v), of a certain two-neuron network is specified as E (v) = − 12 (v12 + 2v1 v2 + 4v22 + v1 ) ,Assuming high-gain neurons, (a) find the weight matrix W and the bias current vector i.

[8]

(b) Determine whether single-layer feedback neural network postulates (symmetry and lack of self-feedback) are fulfilled for W and i computed in part (a). [8] 5. (a) What is the Kohonen layer architure and explain its features. (b) Explain the Kohonen’s learning algorithm. 6. (a) Explain briefly about the counter propagation-training algorithm. (b) Explain the various applications of counter propagation.

[4+4] [4+4] [10] [6]

7. (a) What are the advantages of ART network. Discuss about gain control in ART network. [3+5] (b) Discuss in detail about orienting subsystem in an ART network.

[8]

8. Describe how a neural network may be trained for a pattern recognition task. Illustrate with an example [16] ⋆⋆⋆⋆⋆

1 of 1

Set No. 3

Code No: RR420507

IV B.Tech II Semester Regular Examinations, Apr/May 2007 NEURAL NETWORKS ( Common to Computer Science & Engineering and Electronics & Computer Engineering) Time: 3 hours Max Marks: 80 Answer any FIVE Questions All Questions carry equal marks ⋆⋆⋆⋆⋆ 1. (a) Explain in detail about the single layer artificial neural network with diagram. (b) Explain in detail about the Multi layer artificial neural network with neat diagram. [8+8] 2. Compare the similarities and differences between single layer and multi layer perceptrons and also discuss in what aspects multi layer perceptrons are advantageous over single layer perceptrons. [6+6+4] 3. Describe how a feed forward multi layer neural network may be trained for a function approximation task. Illustrate with an example. [6+10] 4. (a) What are the limitations of Hopfield network? Suggest methods that may overcome these limitations. [4+4] (b) A Hopfield network made up of five neurons, which is required to store the following three fundamental memories: [8] ξ1 = [+1, +1, +1, +1, +1]T ξ2 = [+1, −1, −1, +1, −1]T ξ3 = [−1, +1, −1, +1, +1]T Evaluate the 5-by-5 synaptic weight matrix of the network. 5. Explain the Kohonen’s method of unsupervised learning. Discuss any example as its application. [8+8] 6. Describe the following: (a) Grossberg layer.

[8]

(b) Counter propagation network.

[8]

7. (a) ART network exploits in full one of the inherent advantages of neural computing technique, namely parallel processing ? Explain. [8] (b) Describe the architecture and operation of ART2 network.

[3+5]

8. What are the applications of Kohonen?s networks in image processing and pattern recognition? [16] ⋆⋆⋆⋆⋆

1 of 1

Set No. 4

Code No: RR420507

IV B.Tech II Semester Regular Examinations, Apr/May 2007 NEURAL NETWORKS ( Common to Computer Science & Engineering and Electronics & Computer Engineering) Time: 3 hours Max Marks: 80 Answer any FIVE Questions All Questions carry equal marks ⋆⋆⋆⋆⋆ 1. What are the modes of operation of a Hopfield network? Explain the algorithm for storage of information in a Hopfield network. Similarly explain the recall algorithm. [4+8+4] 2. Briefly discuss about linear separability and the solution for EX-OR problem.Also suggest a network that can solve EX-OR problem. [4+6+6] 3. Implement a backpropagation algorithm to solve EX-OR problem and try the architecture in which there is a hidden layer with three hidden units and the network is fully connected. [8+8] 4. Show how the traveling salesman problem can be solved using the Hopfield model. [16] 5. Discuss how the “Winner-Take-All” in the Kohonen’s layer is implemented and explain the architecture, Also explain the training algorithm. [16] 6. Explain the operation of counter propagation with suitable network model and give the equations for training. [16] 7. Explain the major phases involved in the ART classification process.

[16]

8. Describe how a neural network may be trained for a pattern recognition task. Illustrate with an example [16] ⋆⋆⋆⋆⋆

1 of 1