Deep Learning — DU B. Voc – Software Development Previous Year Question Papers

Uni_que Paper Code : Deep Learning : 8. Voc. (Software T`rne : 3 Hours Ma}ctmum Marks : 90 -fi-a-s--twiro sections. S I. (a) What is deep leaming ? Describe any three major areas where it is commonly applied. (b) Compare and contrast the single - layered model and the multi - layered perceptron model. 5 (c) What is padding in the context of convolutional neural networks (CNNs) ? Explain its purpose and the different types. (d) What is the difference between parameters and tryperparameters in convolutional layers ? List some examples of.each. 5 (e) Differentiate between feed forward neural networks and recurrent neural networks (f) Given a single-layer perceptron with in`put x, weight w, and bias b = 0. Compute # for themeansquarederrorlossL=i(}r-y2)and update the weight w. 5 or®-# 2. (a) Consider a Convolutional Neural Network (CNN) architecture that uses the following notations : (i) CONV-K:N : A,convolutional layer with N filters, each of size K x K. Padding = (ii) POOL-K : A'pooling layer with filter size K x K, stride = K and .padding = 0 (iii) FC-N : A fully connected layer with N neurons. For each layer, calculate the number of weights, the number of biases and the size of the associated feature maps. I.ayer Activation Map Number of `Number of Dimension Weights B iases (b) What is hyperparameter optimization in deep learning ? Explain its importance in model performance and training efficiency. 5 8. (a) What are the key challenges in training deep neural networks (DNNs) ? Describe how these challenges affect the overall performance of the model: 8 (b) Perform the convolution operation on the given 2D inage 'Input' using the provided fflter 'Kernel'. Assume a stride of 1 and no padding (padding = 0). Clearly show all the steps involved in the calculation. 7 4. (a) What are the critical design considerations when developing a recurrent neural network for sequence modeling tasks and what challenges are commonly encountered during its training process? 8 (b) What are the differences between Standard Gradient Descent, Stochas.tic Gradient Descent and Mini - Batch Gradient Descent ? 5. (a') Explain ba`tch normalization in the context of deep neural networks. Describe the key steps involved in performing batch normalization during training a model. (b) Given a single-layer perceptron with. input XL=O, X2=1, WL=0, W2=1, b=O and target outputY = 0 and network output i. Compute 3F'3=andgfortheMeansquaredError loss L=i(9-y2) and update the weight valu es. 7 6. (a) Explain the Sigmoid, Tanh and ReLU activation functions used in deep learning. Describe the main advantages and limitations of each function. 8 (b)¥oohoba±:::e:¥gh¥p:o±;°#:82gis%i¥i¥N:S:Sup::?i:¥=2m¥i fflter and a stride of 2 . Show all steps clearly. Max Pool 7. (a) Explain the architecture and working principle of an autoencoder in deep learning. What are its primary uses and how does it differ from traditional neural networks ? 8