Unit 1

1) List out any four real world applications in deep learning.

2) How deep learning is different from Machine Learning.
3) Is manual feature extraction outdated in Deep Learning?justify.
4) Define Epochs.
5) Define Batch.
6) State the importance of Deep Learning.
7) Define Deep learning.
8) Differentiate deep learning and machine learning.
9) What is machine learning.
Part b
1) Briefly describe the development of deep learning.
2) Derive the update rule for w1 in the regularized neural net loss function which
penalizes based on the square of each weight. Use λ to denote the magic
regularization parameter and also solve Neural Net Simplification Net.
3) Explain the different types of Deep learning architecture.
4) Examine the stochastic gradient types in Deep learning.
5) Briefly describe the development of deep learning.
6) Assume to solve a classification task, in that first train your network on 20 samples.
Training converges, but the training loss is very high, then decide to train this network
on 10,000 examples. Is this approach to fixing the problem correct? If yes, explain the
most likely results of training with 10,000 examples. If not, give a solution to this
problem.
7) Write short notes on (I) CNN (ii) RNN (iii) LSTM
8) Discuss about the optimization algorithm with its type.
9) How optimized algorithm is implemented in neural network?Justify your answer
with its types
10) The following piece of code for forward propagation through a single hidden
layer in a neural network. This layer uses the sigmoid activation. Identify and correct
the error.
import numpy as np
def forward_prop(W, a_prev, b):
z = W*a_prev + b
a = 1/(1+np.exp(-z)) #sigmoid return a

Unit 2
1. Define Loss function.
2. Compare over-fitting and under-fitting.
3. How deep learning is used in real world situation?
4. Differentiate over-fitting and under-fitting.
5. Construct the mathematical representation of neural network.
6. Does deep learning memorize network?
7. How does neural network look?
8. Define layers in DL
9. State importance of binary classification.
Part b
1) Describe the method of deep learning visualisation.
2) Obtain the output of the neuron Y for the network shown using activation function. (i)
binary sigmoidal (ii) bipolar sigmoidal.
3) Derive Backpropogation in representatives of Matrices and Scalar.
4) Describe about the mathematical representation of Neural Network.
5) Discuss about the purpose of different layers in deep learning.
6) Implement AND function using McCulloch-Pitts Neuron (take binary data).
7) Discuss about the types of Gradient descent with its tensors operation.
8) For the network shown in figure, calculate the net input to the neuron

Unit 3
1) If 1D Neural Network is applied in Convolutional Neural Network?Justify.
2) In what cases transfer learning applies in deep learning?
3) Define about the convolutional layer observes?
4) State the applications of RNN.
5) Define Max pooling.
6) Why Recurrent Neural Networks in Deep Learning?
7) State the power of CNN.
8) What each convolution layer sees.
9) Sate real time examples of one dimensional CNN.

Part b
1) 3 34 55
120 12 5
22 50 1

1 2
2) Filters = 3*3 image =
-2 -1
Find the kernals feature map?
Discuss about when to finetune your model in LSTM.
3) How overfitting occurs in RNN?justify your answer with example.
4) Write short notes on (I) one - Dimensional CNN (ii)Multi- Dimensional CNN.
5)
6) 1 0 0 0 0 1 Explain
about 0 1 0 0 1 0 GRU in
Deep learning.
7) 0 0 1 1 0 0 Why
RNN is differ
from 1 0 0 0 1 0 feed
forward
0 1 0 0 1 0 neural
0 0 1 0 1 0 network?
Justify
your answer
with example.
8) Describe
about 1 0 1 RNN and
its -1 0 0 types.
9) 0 1 1 Examine
about when to
For this 6*6 matrix find the convolutional layer with feature finetune
your extraction. model.
10) Provide
brief
descriptions of MIMO Deep Learning Models.
11) The image of twins is confused with each other with (w*h*d)=28*28*1, Number of
kernals is 16, spatial extend of each one must be 5 then stride is 1. if we want padding then it
is around 1.8. find the convolution layer.

Unit 4
1) Define Neural Style Transfer.
2) State the importance about MIMO.
3) List the types of Hyper Parameters.
4) What does a GAN look?
5) Compare about Discriminator and Generator.
6) How Neural Style Transfer works in LSTM?
7) Differentiate between Variational AutoEncoder and AutoEncoder.
8) Differentiate cyclic and acyclic graph.
9) Discuss about GAN.

Part b
1) Write short notes on Generative Adversarial Networks.
2) Write an algorithm for Stochastic Gradient Descent
3) Describe in brief about Variational Auto Encoders.
4) Briefly explain about Graphical Neural Network architecture.
5) Discuss about cyclic and Acyclic graph layers.
6) Write short notes on (I) Bagging (II) ADABoost
7) How LSTM works under text synthesize?justify with an example Dataset.
8) Write short notes on Auto Encoders.
9) Explain about Ensemble methods with example.
10) Explain about image synthesis with Variational Auto Encoders.

Unit 5
1) Define parameter model.
2) List the importance of Radial Bias Function.
3) Define Hippocampal model.
4) State the Baye’s theorm model.
5) State the properties of pattern associates.
6) List the advantages and disadvantages of parameter model.

Part b
1) Explain in detail about Hopfield network with examples.
2) Discuss about Hebbian algorithm.
3)

4) Explain about Radialbias function in Neural Network.

5) Explain the maximum Posterior parameter works in deep learning.
6) Write short notes on Boltzmann Learning.
7)

This is the Truth Table of AND Gate using bipolar sigmoidal function.
There are 4 training samples, so there will be 4 iterations. Also, the activation function used
here is Bipolar Sigmoidal Function so the range is [-1,1]. find the Hebbian model.
8) Explain in detail about Hopfield network with examples.
9) Using Boltzmann Learning explain how real time applications are working.
10) Explain the Posterior parameter works in deep learning.
11)

12) Write brief notes on (i) Maximum a posterior approach (ii) Posterior approach