FACULTY OF ENGINEERING
SCHOOL OF INFORMATION, SECURITY AND DATA SCIENCE
B. Tech IT – VII-Sem
Session: JULY-NOV (2024)
DEEP LEARNING (IT4141)
EXTRA CLASS ASSIGNMENT- I
Q 1. Define the terms 'bias' and 'variance' with respect to a dataset.
Recognize a suitable threshold value to realize a two input NAND gate using
Q 2.
McCulloch-Pits neuron model if the weight values are w1 = -1 and w2 = -1. Assume
that the input and output system have binary variables only.
Q 3. Define briefly vanishing gradient problem in Gradient Descent Algorithm.
Compare momentum based gradient descent algorithm with the vanilla gradient
Q 4.
version. Also discuss the change in the equations.
Compare ReLU activation function with Sigmoid. Also for given figure below,
calculate the gradients ∂f/∂x, ∂f/∂y, ∂f/∂z using Gradient Descent Algorithm.
Q 5.
In face recognition, the goal is to identify or verify a person’s identity based on their
facial features. The input data consists of images of faces, which are typically high-
dimensional (e.g., 64x64 pixels means 4,096 dimensions per image). Directly using
this high-dimensional data for classification can be inefficient and may lead to
Q6.
overfitting. Suggest what will be your preffered choice of autoencoder in the neural
network and why?
Calculate the new transformed points corresponding to the first principle component
Q7. for the given data points (4,2) and (3,6) investigating the eigen values and vectors
using Principle Component Analysis.
Train the network shown in the figure with X1=0.05; w1 = 0.15, w2 = 0.25, w3 =
0.40, w4 = 0.50 b = 0.35 and lr =0.5.
(Assume Sigmoid as activation function and mean square error)
Q 8.
Assignment submission due date: 21/11/2024 in 2nd Extra class
