Scribd
Upload
18 views

Intro To CNN

This document discusses and compares convolutional neural networks (CNNs) and feedforward neural networks (FFNNs). CNNs have sparse connections and fewer learnable parameters compared to FFNNs, which have dense connections and more parameters. This helps reduce overfitting in deep neural networks. The document then demonstrates how a CNN uses convolutional layers with filters, activation maps, depth, stride, and padding to analyze images like the MNIST dataset with fewer parameters than an FFNN. The CNN applies filters to extract features from the input image to generate an output volume.

Uploaded by

Mehak Smagh
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
18 views

Intro To CNN

This document discusses and compares convolutional neural networks (CNNs) and feedforward neural networks (FFNNs). CNNs have sparse connections and fewer learnable parameters compared to FFNNs, which have dense connections and more parameters. This helps reduce overfitting in deep neural networks. The document then demonstrates how a CNN uses convolutional layers with filters, activation maps, depth, stride, and padding to analyze images like the MNIST dataset with fewer parameters than an FFNN. The CNN applies filters to extract features from the input image to generate an output volume.

Uploaded by

Mehak Smagh
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 17

Convolutional Neural Network (CNN)

• Why CNN?
• Deep neural networks
– Increased no of parameters
– Overfitting
• Can we reduce number of parameters?
– CNN
Difference between FFNN and CNN

FFNN CNN
Dense Connections Sparse connections
No. of learnable parameters - No. of learnable parameters -
High Low
Learnable parameters
MNIST dataset
• FFNN
• 1st layer :
– (256 hidden units and input size=28×28)
• 2nd layer:
– (10 units at 2nd layer)
Learnable parameters
MNIST dataset
• FFNN
• 1st layer :
– (256 hidden units and input size=28×28)
– (784*256)+256=200960
• 2nd layer:
– (10 units at 2nd layer)
– (256*10)+10=2570
CNN: Convolutional layer

• In mathematics convolution is a mathematical


operation on two functions (f and g) that
produces a third function a.

a: https://en.wikipedia.org/wiki/Convolution
CNN: Convolutional layer

• Filter or kernel
– Extract features from images
– Moves over the input data from left to right top to bottom,
computes dot product with sub-region of input data
• Activation map or feature map
– Output volume formed by sliding the filter over the
image
• Depth
– Number of filters
• Stride
– Main objective is to produce smaller output
volumes,
– If s=2, the filter will shift by 2 pixels as it
convolves around input volume.
– Size of the image keeps on reducing as we increase
the stride value.
• Padding
– Padding the input image with zero across it.
CNN
• Input layer
• Convolutional layer
• Fully connected layer
• Output layer
Convolutional layer

Consider the following 6x6 image and the following 3x3 filter

3 0 1 2 7 4
1 5 8 9 3 1 1 0 -1
2 7 2 5 1 3 1 0 -1
0 1 3 1 7 8
1 0 -1
4 2 1 6 2 8
2 4 5 2 3 9 KERNEL

INPUT
Convolutional layer

3 0 1 2 7 14 0 -1 3 0 -1
1 5 8 9* 3 11 0 -1 = 1 0 -8
2 7 2 5 1 13 0 -1 2 0 -2
0 1 3 1 7 8
4 2 1 6 2 8
2 4 5 =>23+0+(-1)+1+0+(-8)+2+0+(-2)=
3 9 -5
Convolutional layer

Consider the following 6x6 image and the following 3x3 filter

3 0 1 2 7 4
1 5 8 9 3 1 1 0 -1
2 7 2 5 1 3 1 0 -1
0 1 3 1 7 8
1 0 -1
4 2 1 6 2 8
2 4 5 2 3 9
0 1 2 1 0 -1 0 0 -2
5 8 9 * 1 0 -1 = 5 0 -9
7 2 5 1 0 -1 7 0 -5

=> 0+0+(-2)+5+0+(-9)+7+0+(-5)= -4
Similarly after applying the filter to the complete image, we obtain the following results in
the form of a 4x4 matrix

5 -4 0 8
-10 -2 2 3
0 -2 -4 -7
-3 -2 -3 -16
Computing output Volume [W2×H2×D2]

You might also like

pFad - Phonifier reborn

Pfad - The Proxy pFad of © 2024 Garber Painting. All rights reserved.

Note: This service is not intended for secure transactions such as banking, social media, email, or purchasing. Use at your own risk. We assume no liability whatsoever for broken pages.


Alternative Proxies:

Alternative Proxy

pFad Proxy

pFad v3 Proxy

pFad v4 Proxy