Data mining#ANN

Which of the following statement about ANN is false?

• ANN is preferred when Input is high-dimensional discrete or real-valued

• ANN is preferred when less training time is desired

• ANN is preferred when input data contain noisy data

• ANN is preferred when Output is discrete or real valued

In the perceptron training rule, weights are updated as follows. 
 $W_i\ =\ W_i+\ \Delta W_{i\ }$  Where  $\Delta W_{i\ }\ =\$  -----------?

•  $\eta\left(t\ -o\right)x_{i\ }$  

•  $\left(t\ -\ o\right)x_i$  

•  $\eta\left(t\ -\ o\right)$  

•  $\eta\left(t-o\right)w_i$  

Perceptron training rule guaranteed to succeed if

1) Training examples are linearly separable

2) Even when training data contains noise

3) Even when training data not linearly seperable

4) Sufficiently small learning rate η

• 1 and 2 only

• 1 only

• 1 and 4 only

• 4 only

Which of the following statement about "Linear unit training rule uses gradient descent" is false

• Guaranteed to converge to hypothesis with minimum squared error

• Gradient descent works when given sufficiently small learning rate η

• Gradient descent works when training data contains noise

• Gradient descent works only when training examples are linearly separable

What is the derivative of the sigmoid function $\sigma\left(x\right)\ =\ \frac{1}{\left(1\ +\ e^{-x}\right)}$ ?

•  $\sigma\left(x\right)\left(1\ -\ \sigma\left(x\right)\right)$  

•  $\sigma\left(x\right)\left(1-\sigma\left(x\right)^2\right)$  

•  $\left(1\ -\sigma\left(x\right)\right)$  

•  $\frac{\sigma\left(x\right)}{\left(1-\sigma\left(x\right)\right)}$  

Consider a multilayer feedforward network .  The error function is  $\frac{1}{2}\sum_{d\ \in D}^{ }\left(t_d\ -o_d\right)^2$  where D is the set of training examples  $t$   is the target value and  $o$   is the outout value. Assume backpropagation algorithm is used. What will be  $\frac{\partial E}{\partial W_i}$  

•  $-\sum_{d\ \in D}^{ }\left(t_d-o_d\right)o_d\left(1\ -\ o_d\right)x_{i,d}$  

•  $-\sum_{d\in D}^{ }\left(t_{d\ }-o_d\right)x_{i,d}$  

•  $-\sum_{d\in D}^{ }o_d\left(1-o_d\right)x_{i,d}$  

•  $\sum_{d\in D}^{ }\left(t_d-o_d\right)o_d\left(1-o_d\right)x_{i,d}$  

Consider the following Multilayer feedforward network with initial values as follows . Find the output at the fifth unit.

• 0.332

• 0.525

• 0.474

• -0.7

Consider the following Multilayer feedforward network with initial values as follows . Find the output at the fourth unit.

• 0.332

• 0.525

• 0.474

• -0.7

Consider the following Multilayer feedforward network with initial values as follows . Find the output at the sixth unit.

• 0.332

• 0.525

• 0.474

• -0.7

Consider the following Multilayer feedforward network with initial values as follows . Find the error at the sixth  $\left(\delta_6\right)unit$   

• 0.1311

• 0.0065

• -0.1311

• -0.7