You are on page 1of 12

Webinar: Machine Learning with

Python

2D CNN ARCHITECTURE
IMPLEMENTATION
IN KERAS
Speaker
Dr. M. Kalaiselvi Geetha
Professor
Department of Computer Science and
Engineering
Annamalai University
Application: CIFAR-10 Classification

Creating
Formation
the Compiling
of the
2D CNN the Model
Dataset
Model

Airplane
Automobile
Bird
Cat
Deer Training
Dog Testing and
Validation
Frog
Horse
Ship
Truck

2/12
Dr. M. Kalaiselvi Geetha, Professor of CSE, FEAT, Annamalai University
Samples of CIFAR-10 Data Sets

3/12
Dr. M. Kalaiselvi Geetha, Professor of CSE, FEAT, Annamalai University
Implementation of 2D CNN using Keras

No. of Trainable Parameters/weights:


FC1: FC2: SOFTMAX LAYER:
CONV1: CONV2: CONV3:
(500 x 4480) + 500 (250 x 500) + 250 (10 x 250) + 10
((3 x (3x3)) +1)x25 ((25 x(3x3) +1) x50 ((50x(3x3) +1)x70
2240500 125250 2510
700 11300 31570
Convolution: 3 x 3
Max Pooling: 2 x 2
Total Trainable Parameters: 24,11,830

4/12
Dr. M. Kalaiselvi Geetha, Professor of CSE, FEAT, Annamalai University
Formation of Training , Validation, and Test Data Sets OUR SOLUTIONS

During training 20% of training data


is assigned for validation using
validation_split parameter in fit
method.

5/12
Dr. M. Kalaiselvi Geetha, Professor of CSE, FEAT, Annamalai University
Model Architecture of 2D CNN using Keras

6/12
Dr. M. Kalaiselvi Geetha, Professor of CSE, FEAT, Annamalai University
Training and Validation

7/12
Dr. M. Kalaiselvi Geetha, Professor of CSE, FEAT, Annamalai University
Testing
OUR SOLUTIONS

8/12
Dr. M. Kalaiselvi Geetha, Professor of CSE, FEAT, Annamalai University
Testing

9/12
Testing

10/12
Dr. M. Kalaiselvi Geetha, Professor of CSE, FEAT, Annamalai University
References
OUR SOLUTIONS

1. Francois Chollet, Deep Learning with Python, Manning Publications, 2018.

2. Python Deep Learning library Keras documentation: https://keras.io

3. Python Package Index: https://pypi.org

4. URL: https://towardsdatascience.com/

11/12
Dr. M. Kalaiselvi Geetha, Professor of CSE, FEAT, Annamalai University
QUERIES?

THANK YOU FOR LISTENING !

12/12

You might also like