Architecture of processing of Deep feed forward neural networks.

working of a Linearly Seperable AND gate using feed forward neural networks
challenges faced by a feed forward neural network during training. how can they be overcomed.
how XOR problem will be solved using deep neural network
Feed Forward and Back Propagation Neural Network.
architectural components of Deep Feed Forward Networks
working of Batch, stochastic and mini batch gradient descent using a simple usecase.
significance of Loss function in Gradient descent? Which common loss functions used in gradient
descent.
single-layer neural network with weight w and bias b, derive the weight and bias update equations
using Stochastic Gradient Descent for a given error function Cw,b).
brief various gradient descent algorithms
effectiveness of different gradient descent optimization strategies in training deep neural networks.
choice of activation functions affect gradient descent in neural networks?
better choice of activation function
Chain rule in back propagation with a neat diagram.
***working of back propagation in a multi-layered perceptron. With example problem
working of different activation functions in back propagation.
the neurons have a sigmoid activation function, perform a forward pass and a backward
pass on the network. Assume that the actual output of y is 1 and learning rate is 1?
neural network with mean squared error loss, derive the gradient descent update rule for the
weights of the output layer.
Why can't a single-layer perceptron learn the XOR function, and how does backpropagation
enable a multi-layer perceptron to solve it?
L1 and L2 regularization in detail, including their mathematical equations, and describe how
each technique affects weight updates during training in neural networks?
descriptive reasons on why regularization with constraints is chosen than by norm penalties.
L1 and L2 regularization help prevent overfitting in neural networks?
various Parameter Norm Penalties
importance of regularization in Deep Learning models. Eloborate various approaches followed
brief Norm Penalties as Constrained Optimization

different types of convolution functions and their roles in deep learning models.
importance of separable convolution in deep learning and compare it with standard convolution
with example.
Different Types of Convolution: Standard, Strided, and Tiled Convolution with Examples
Role of Tiled Convolution in Optimizing Deep Learning Models
number of multiplications required for a convolution operation on a 16×16 input image using a 3×3
kernel with a stride of 1. Extend this to show how computational cost changes with different
strides.
A 5×5 image is convolved with a 2×2 filter using tiling, where each tile is of size 2×2.
Compute the number of tiles required and explain how tiled convolution reduces memory usage.
impact of overfitting in CNNs. the techniques used to prevent it?
architecture of a Convolutional Neural Network (CNN) and describe its key components.
role of convolutional layers in CNNs. How do filter size, stride, and padding impact feature
extraction?
Explain how the ReLU activation function enhances the learning capability of CNNs. Why is ReLU
preferred over sigmoid or tanh in deep networks?
Analyze the importance of a convolutional kernal in CNN? Examine various convolutional kernals
used in CNN
feature maps for convolution of the image [1 2 3 4;4 3 2 1;1 3 5 7;2 4 6 8] with a 5x5 average filter
mask. use padding values to show the effect of padding
Explain how the combination of ReLU, pooling, and padding contributes to the efficiency and
accuracy of CNNs.
the necessity of pooling layers in CNNs. Can CNNs work effectively without pooling? Discuss
alternative approaches such as strided convolutions.
what is Padding ? why is Padding required? Write on various types of Padding?
how random initialization of CNN filters leads to frequency selectivity and translation
invariance.
role of hand-designed kernels in CNNs. Provide examples of their application and compare
their advantages and disadvantages with learned kernels.
13 pixel values (1,2), (2,2) (3,1) (3,4) (3,7) (4,2) (4,1) (4,4) (4,5) (5,1) (5,5) (6,2) (6,4) for example,
apply 3-Means algorithm and show why unsupervised algorithms are preferred in convolution
filters?
role of the primary visual cortex (V1) in visual processing and how it is replicated in
Convolutional Neural Networks (CNNs).
differences between the Human Visual System (HVS) and Convolutional Neural Networks
(CNNs).
Gabor functions relate to visual processing in V1 neurons.

components of an LSTM network and how they work together to store and update information
the input, forget, and output gates in an LSTM contribute to its performance.
training process of an LSTM and how it differs from traditional RNNs in terms of handling gradients.
strengths and weaknesses of LSTM compared to other sequence models, such as GRU and standard
RNN.
LSTMs are used in applications like speech recognition and language translation.
how LSTMs handle long-term dependencies in sequences
the role of the hidden state in RNNs. How does it help in learning sequential patterns?
Compare RNN, LSTM, and GRU in terms of architecture, performance
 the impact of different activation functions ReLU, Tanh, Sigmoid in RNN and GRU models.
the effectiveness of GRUs in time-series forecasting compared to traditional RNNs. When should
GRUs be preferred?
Compare architectures of LSTM, and GRU in terms of training time, and performance on sequential
data.
the working principle of RNN with an illustrative example
sequence learning problem for the task of predicting the polarity of a
review
Develop the steps in Backpropagation Through Time (BPTT) algorithm applied to capture temporal
dependencies in sequential data?
idea of unfolding a recursive or recurrent computation into a
computational graph that has a repetitive structure. Draw the
structures.

Backpropagation Through Time (BPTT) and its role in loss computation and weight updates.
Explain the working principle of sequence to sequence model

Explain Encode-Decoder sequence to sequence models. List the pitfalls

of them
Compare Recursive Neural Networks, Recurrent Neural Networks and
Bidirectional Recurrent Neural Networks based on design, functioning
and performance complexity
the working of deep recurrent neural network for machine language
translation
Discuss the merits and demerits of using recursive neural network for
natural language processing.