You are on page 1of 2

187105, Ajeet Singh

Deep CNN to solve handwritten recognition problem for A to Z


and 0 to 9 letters

Introduction
Deep convolutional neural networks (CNN or DCNN) are the type most commonly
used to identify patterns in images and video. DCNNs have evolved from
traditional artificial neural networks, using a three-dimensional neural pattern inspired
by the visual cortex of animals.

Process
It includes several steps for calculating accuracies
1.Data Processing- used the data downloaded from kaggle for lab assignment and took 50
images for every digit and letter
2.Building the Model - I used the Keras API to build the model hence I have a Tensorflow
background
3. Compiling and fitting the Model - Here we get pretty high accuracy with just 10 epochs.
Since the dataset doesn’t need heavy computational power we can work around with the
number of epochs , loss function and metrics.
4. Model Evaluation - After giving all the parameters we run our model for each activation
functions

Observations

Number of epochs
Number of epochs is the number of times the whole training data is shown to the network
while training.
Increase the number of epochs until the validation accuracy starts decreasing even when
training accuracy is increasing(overfitting).
Here we are using 10 epochs

Batch size
Mini batch size is the number of sub samples given to the network after which parameter
update happens.A good default for batch size might be 32. Also try 32, 64, 128, 256, and so
on.We have used the batch size of 32

Learning Rate
The learning rate defines how quickly a network updates its parameters.
Low learning rate slows down the learning process but converges smoothly. Larger learning
rate speeds up the learning but may not converge.
Here we are using learning rate as 0.01

Results
After calculating the results for each of the activation functions for 10 epoch and
observed their accuracies and then plotted them on the graph and results are shown
below.

As we can see in the graph tanh has given more accuracy compared to other
activation functions And sigmoid has given very less accuracy close to 0.1. And relu
,tanh ,elu are also very close if seen from a high level.

You might also like