You are on page 1of 2

QUESTION 2 – Convolution Neural Network

Q2 A – Consider the following CNN architecture:

 Input layer: 5x5x1


 Convolution layer with rectified linear units (ReLU): 2 filters of size 3x3x1,
applied with zero-padding of P=1 and Stride S=1. The filter weights are given
below.

0.5 0 0
0 0 0
0 0 0.5

1 0 -1
0 0 0
1 0 0

 Max Pooling layer: size = 3x3, zero-padding P=0, and stride S = 2;


 Fully Connected layer: Two neurons in the fully connected layer. All weights
corresponding to the first neuron are 1, and for the second neuton the
weights are 2 for all inputs.

Given the following input image compute the values at all the intermediate layers.

Input Image =

3 1 10 5 3
5 3 5 4 2
2 1 1 2 1
5 2 1 1 6
1 9 2 4 8
Registeration # __________________________

Q2 B – Find the total number of parameters/weights and the memory required (in
Bytes) to hold all the intermediate hidden layers (including the input and final
output layer) in the following network. All convolution filters have size 3x3, stride 1
and pad 1, and all pool layers size is 2x2, stride 2.

Input layer: 224x224x3


Conv layer with 64 filters
Conv layer with 64 filters
Pool Layer
Conv layer with 128 filters
Conv layer with 128 filters
Pool Layer
Conv layer with 256 filters
Conv layer with 256 filters
Conv layer with 256 filters
Pool Layer
Conv layer with 512 filters
Conv layer with 512 filters
Conv layer with 512 filters
Pool layer
Conv layer with 512 filters
Conv layer with 512 filters
Conv layer with 512 filters
Pool layer
FC – 4096 neurons
FC – 4096 neurons
FC – 1000 neurons

P a g e 2 of 2

You might also like