Convolutional Neural Network

Program implements a convolutional neural network for classifying images of numbers in the MNIST dataset as either even or odd using GPU framework.

1. Contructing CNN: MNIST dataset is split into training, validation, and testing subsets with 55000, 10000, and 5000 examples, respectively. Digit labels are converted to even or odd using 0 and 1. Also plotting training and validation loss and accuracy as a function of epochs.
2. Hyper-parameters Tuning: Evaluating different variations of basic network by changing:
   • network architecture
   • receptive field using dilation
   • stride
   • optimizer
   • loss function
   • dropout
   • learning rate
   • number of epochs
   • weight initializers
   • batch normalization
   • layer normalization
3. Inference: Taking images with handwritten digits as input and performing image preprocessing like resizing, converting to grayscale and then to binary image using threshold. Classifying input images into odd/even labels.

Program was implemented using Python, TensorFlow, Keras, and OpenCV. Refer the report for further implementation details and instructions to run the code: View Report

Results:

1. Contructed CNN for training:

2. After Hyper-parameter Tuning:

Parameters	Values
Dilation	1
Stride	1
Optimizer	Adam
Loss function	Binary cross entropy
Dropout	0.3
Learning rate	0.001
Weight Initializer	He normal
Epochs	15
Batch Normalization	Added layers after convolution and max pooling layers
Layer Normalization	Added layers after convolution layers

3. Inference with handwritten images: