prachip23 / classification-on-mnist

Implementation and evaluation of classification algorithms- logistic regression, single hidden layer neural network, convolutional neural network on MNIST dataset

Classification on MNIST

1. Implemented digit classification of MNIST images by Logistic Regression, Multilayer Perceptron Network and Convolutional Neural Network using TensorFlow Library.
2. Tuned the hyper parameters - used different optimizers, varied learning rate, varied number of neurons and number of layers to improve the efficiency. Compared the efficiency for all three methods.
3. Verified the ‘No Lunch Theorem’ by testing the trained model on USPS dataset The Report contains detailed results.