In my experiment, I train a multilayer CNN
for street view house numbers recognition and check the accuracy of test data.
The coding is done in python using Tensorflow, a powerful library for
implementation and training deep neural networks. The central unit of data in
TensorFlow is the tensor. A tensor consists of a set of primitive values shaped
into an array of any number of dimensions. A tensor’s rank is its number of
dimensions. 20 Along with TensorFlow used some other library function such as
Numpy, Mathplotlib, SciPy etc.
Firstly, as I have technical resource
limitation I perform my analysis only using the train and test dataset. And
omit extra dataset which is 2.7GB. Secondly, to make the analysis simpler I
find and delete all those data points which have more than 5 digits in the
image. For the implementation, I randomly shuffle valid dataset I have used the
pickle file svhn_multi which I created by preprocessing the data from the
original SVHN dataset. Then used the pickle file and train a 7-layer Convoluted
Neural Network. Finally, I cast off the test data to check for accuracy of the
trained model to detect number from street house number image.
the very beginning of my experiment, first convolution layer I used 16 feature
maps with 5×5 filters, and originate 28x28x16 output. A few ReLU layers are
also added after each layer to add more non-linearity to the decision-making
process. After first sub-sampling the output size decrease in 14x14x10. The
second convolution has 512 feature maps with 5×5 filters and produces 10x10x32
output. In this moment applied sub-sampling second time and shrink the output
size to 5x5x32. Finally, the third convolution has 2048 feature maps with same
filter size. It is mentionable that the stride size =1 in my experiment along
with this zero padding also used here. During my experiment, I used dropout
technique to reduce the overfitting. Finally, the last layer is SoftMax
regression layer. Weights are initialized randomly using Xavier initialization
which keeps the weights in the right range. It automatically scales the
initialization based on the number of output and input neurons. Now I train the
network and log the accuracy, loss and validation accuracy in steps of 500.
Initially, I used a static learning rate of
0.01 but later switched to exponential decay learning rate with an initial
learning rate of 0.05 which decays every 10000 steps with a base of 0.95. To
minimize the loss, I used Adagrad Optimizer. When I reached a satisfactory
accuracy level for the test dataset then stop the learning and save the
hyperparameters in the cnn_multi checkpoint file. When I need to perform the detection,
it will load that time without train the model again.
Initially, the model produced an accuracy of
89% with just 15000 steps. It’s a great starting point and certainly, after a
few hours of training the accuracy will reach my benchmark of 90%. However, I added
some simple improvements to further increase the accuracy of few number of
learning steps. First, added a dropout layer after the third convolution layer
just before fully connected layer. This allows the network to become more
robust and prevents overfitting. Secondly, introduced exponential decay to
learning rate instead of keeping it constant. This helps the network to take
bigger steps at first so that it learns fast but over time as we move closer to
the global minimum, take smaller noisier steps. With these changes, the model
is now able to produce an accuracy of 92.9% on the test set with 15000 steps.
Since there are a large training set and about 13068 images in the test set,
there is a chance of more improvement if we trained the model for a longer duration.