# (Aktu Btech) Application of Soft Computing Important Unit-2 Neural Network-II (Back Propagation Network)

In the B.Tech AKTU Quantum Book, learn about the practical Application of Soft Computing. Discover crucial applications, frequently asked questions, and essential notes for mastering this cutting-edge technology. Unit-2 Neural Network-II (Back Propagation Network)

```Dudes 🤔.. You want more useful details regarding this subject. Please keep in mind this as well.

Important Questions For Application of Soft Computing:
*Quantum                *B.tech-Syllabus
*Circulars                 *B.tech AKTU RESULT
* Btech 3rd Year     * Aktu Solved Question Paper```

## Q1. What is the multilayer perceptron model? Explain it.

Ans.

• 1. A multilayer perceptron is a feed forward artificial neural network class.
• 2. A multilayer perceptron model includes three layers: an input layer, an output layer, and a layer in between that is not directly connected to either the input or the output and is hence referred to as the hidden layer.
• 3. We utilise sigmoidal or squashed-S function for perceptrons in the input layer and linear transfer function for perceptrons in the hidden and output layers.
• 4. Because the input layer’s purpose is to distribute the values it receives to the next layer, it does not conduct a weighted sum or threshold.
• 5. The input-output mapping of multilayer perceptron is shown in Fig. and is represented by
• 6. Multilayer perceptron does not increase computational power over a single layer neural network unless there is a non-linear activation function between layers.

Ans.

## Q3. What are the back propagation learning methods ?

Ans. Learning methods of back propagation

• 1. Static back propagation :
• a. It is a back propagation network that generates a mapping of a static input to a static output.
• b. It can be used to address static classification problems such as optical character recognition.
• c. In static back propagation, the mapping is fast.
• 2. Recurrent back propagation:
• a. Recurrent back propagation is continued until a fixed value is reached.
• b. The mistake is then computed and propagated backward.
• c. In recurrent back propagation, it is non-static.

## Q4. Write down the advantages and disadvantages of back propagation networks.

Ans. Advantage of back propagation networks/algorithm:

• 1. It is quick, straightforward, and simple to programme.
• 2. There are no tuning parameters (except for the number of input).
• 3. There is a batch weight update that gives a smoothing impact on the weight of correction terms.
• 4. The computing time is lowered if the weight specified at the start is minimal.

• 1. The actual performance of back propagation on a specific problem depends explicitly on the input data.
• 2. Back propagation is susceptible to noise and outliers.
• 3. Instead of a mini-batch, a fully matrix-based technique is employed for back propagation.
• 4. Once a network has learned one set of weights, every additional learning results in catastrophic forgetting.

## Q5. Discuss the selection of various parameters in BPN.

Ans. Selection of various parameters in BPN (Back Propagation Network):

1. Number of hidden nodes:

• i. The leading criterion is to select the fewest nodes possible without impairing network performance, so that the memory required to store the weights is kept to a minimum.
• ii. When the number of concealed nodes equals the number of training patterns, learning may be accelerated.
• In such instances, the Back Propagation Network (BPN) recalls training patterns but loses all generalisation skills.
• iv. Hence, as far as generalization is concerned, the number of hidden nodes should be small compared to the number of training patterns (say 10:1).

2. Momentum coefficient (α):

• i The another method of reducing the training time is the use of momentum factor because it enhances the training process.
• ii. The momentum also overcomes the effect of local minima.
• iii. It will carry a weight change process through one or local minima and get it into global minima.

3. Sigmoidal gain (𝝺):

• i. When the weights become large and force the neuron to operate in a region where sigmoidal function is very flat, a better method of coping with network paralysis is to adjust the sigmoidal gain.
• ii. By decreasing this scaling factor, we effectively spread out sigmoidal function on wide range so that training proceeds faster.

4. Local minima:

• i. One of the most practical solutions is to introduce a shock that causes all weights to vary by specific or random amounts.
• ii. If this fails, the remedy is to re-randomize the weights and begin training again.
• iii. Simulated annealing is employed to keep training going until a local minima is reached.
• iv. Simulated annealing is then paused, and BPN proceeds until a global minimum is reached.
• v. In most cases, this two-stage procedure requires only a few simulated annealing cycles.

5. Learning coefficient (η):

• i. If the learning coefficient is negative, the change in weight vector position will deviate from the optimum weight vector position.
• ii. If the learning coefficient is zero, there is no learning; thus, the learning coefficient must be positive.
• iii. If the learning coefficient is more than one, the weight vector will overshoot and fluctuate from its optimal location.
• iv. As a result, the learning coefficient must be between 0 and 1.

## Q6. Discuss how learning rate coefficient affects the back propagation training.

Ans.

• 1. The learning rate coefficient influences the rate of convergence by determining the amount of the weight adjustments made at each iteration.
• 2. A poor choice of coefficient can lead to a failure of convergence.
• 3. For the best outcomes, we should keep the coefficient constant across all iterations.
• 4. If the learning rate coefficient is too large, the search path will oscillate and converges more slowly than a direct descent as shown in Fig.(a).
• 5. If the coefficient is too small, the descent will progress in small steps significantly increasing the time to converge as shown in Fig.(b).